Research Article For reprint orders, please contact: [email protected]
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
Aim: Nine months of isoniazid (9INH) is the gold standard for treatment of latent tuberculosis infection (LTBI). This paper compares the effectiveness of 9 months of isoniazid with 4 months of transitional rifampin (9H4R) to alternative therapies, including 9INH, 6 months of isoniazid (6INH) and 6 months of isoniazid with 4 months of transitional rifampin (6H4R), for treatment of LTBI. Materials & methods: Using an ethnically diverse clinic sample of 552 patients given treatment for LTBI with 9H4R, we use multilevel selection models to examine the adjusted comparative effectiveness of the regimens among ethnic groups that feature distinct genetic predispositions to side effects on INH. For unadjusted/absolute effectiveness, we simulated cost– effectiveness ratios for 4 months of rifampin (4RIF) and compared with bootstrapped confidence intervals for the alternative therapies. Results: There are variations in the comparative effectiveness across ethnic groups, with the most notable differences for 9H4R. For unadjusted/absolute effectiveness, 4RIF presents the greatest net benefit for US born black and African patients. For all other ethnic groups, 6H4R was the most effective. Conclusion: Patient ethnicity affects tolerance to INH. 9H4R was the most effective LTBI treatment for all ethnicities. However, this result heavily depends on whether adjustments are made for self-selection.
Kyle R Fluegge*,1,2,3,4 & Brian E Roe1 Department of Agricultural, Environmental & Development Economics, Ohio State University, 2120 Fyffe Road, Columbus, OH 43210, USA 2 Division of Epidemiology, College of Public Health, Ohio State University, 2120 Fyffe Road, Columbus, OH 43210, USA 3 Institute for Health & Environmental Research, Columbus, OH 43220, USA 4 Department of Epidemiology & Biostatistics, Case Western Reserve University, 2103 Cornell Road, Cleveland, OH 44106, USA *Author for correspondence: [email protected] 1
Keywords: comparative effectiveness • cost–effectiveness • ethnicity • isoniazid • latent tuberculosis infection • Ohio • rifampin
Tuberculosis (TB) remains a major global health concern. In 2010, there were 8.8 million worldwide incident cases, from which more than a million individuals die annually [1] . Furthermore, the Centers for Disease Control and Prevention (CDC) estimates that a third of the world’s population is infected with Mycobacterium tuberculosis [2] . This constitutes a large reservoir of patients with latent tuberculosis infection (LTBI), some of whom will develop active disease if not treated [3] . Two LTBI treatment regimens have received the most attention: 9 months of daily isoniazid (9INH) and 4 months of daily rifampin (4RIF). 9INH is less expensive. However, the 4RIF treatment is shorter and causes fewer side effects, which often results in greater rates of respondent completion for 4RIF. Side effects may be, in part, geneti-
10.2217/CER.15.3 © 2015 Future Medicine Ltd
cally determined, which gives rise to questions of comparative effectiveness for patient groups with substantial genetic variation. In this paper, we study a clinically derived LTBI treatment regimen: 9INH with transitional 4RIF (9H4R) if INH is not well tolerated. As a result, we are able to more fully assess outcomes for patients who would have otherwise had 9INH therapy terminated as well as comparing outcomes to several alternative treatment definitions including 9INH alone, 6 months of isoniazid (6INH) or 6INH with transitional 4RIF (6H4R). The questions of relevance in this paper are does the relative effectiveness of 9H4R, 9INH, 6INH or 6H4R differ across patient ethnicity in a predictable fashion and is any one or more of these regimens more or less cost effective than a simulated regimen of 4RIF alone?
J. Comp. Eff. Res. (2015) 4(3), 239–257
part of
ISSN 2042-6305
239
Research Article Fluegge & Roe Multiple studies consider the relative cost–effectiveness (CE) of 4RIF, 9INH and 6INH. Most consist of CE analyses done in the context of randomized-controlled trials (RCTs) [4–8] , whereas others use retro spective clinical data [9] . RCTs are widely considered the superior methodology to assess treatment outcomes and their costs. The process of randomization in RCTs is thought to prevent bias that occurs with participant self-selection. In other words, randomization of patients to treatment removes patient heterogeneity that may be correlated with treatment outcomes. However, randomization may be unable to eliminate bias due to patients who are unwilling to participate (i.e., the unenrolled). In this case, researchers are faced with potential systematic differences between the trial and general patient population. Bias may also result from unmeasured and unpredicted effects of treatment on the outcomes of interest, choice of randomizing variables that are correlated with treatment benefits [10] or complete omission of randomization when it is needed. In this paper we describe a single-arm comparative effectiveness methodology [11,12] with ethnic subgroup analyses to test the effectiveness of 9H4R, 9INH, 6INH and 6H4R, and examine the absolute effectiveness of these regimens compared with a simulation of care with only 4RIF. These inquiries are shown in Figures 1 & 2. By a single-arm, we refer to a design wherein all patients are started on the same medication (INH), but may have switched to rifampin as a result of side effects on INH (the 9H4R and 6H4R definitions). We use a sample of 552 clinic patients with medically confirmed diagnoses of LTBI that began INH. Adherence is determined by the treatment definitions (i.e., 9H4R, 9INH, 6INH and 6H4R). To assess the comparative effectiveness, we use a series of multilevel regression models to analyze patient selection into treatment, patient attrition from treatment and accumulated costs of treatment, where the multiple levels involve treatment months, individual patients and a grouping variable based on patient ethnicity. All models control for patient ethnicity because different ethnic groups acetylate (metabolize) isoniazid differently [13,14] . Asian patients, who are expected to best tolerate 9INH [15,16] , serve as the reference group. To analyze absolute effectiveness, we compute bootstrapped confidence intervals for the CE ratios of 9H4R, 9INH, 6INH and 6H4R for each ethnic group. Simulated CE ratios of 4RIF are compared with these confidence intervals to determine if 4RIF improves treatment effectiveness more than the various INH-based therapies. The contributions of this work inform the literature in several ways. First, we are the first to investigate the
240
J. Comp. Eff. Res. (2015) 4(3)
potential of transitional LTBI therapy (i.e., 9H4R, 6H4R), which has been suggested as a potentially effective modified protocol for patients experiencing side effects on INH [17] . Second, we incorporate selection models into the comparative effectiveness analyses to address the dual nature of selection; that is, selection into treatment and attrition from treatment, which are largely disregarded in RCTs. Third, we use a comparative effectiveness modeling framework with random effects (intercepts and slopes) to specifically model net benefit differences (i.e., effectiveness) between ethnic groups. Last, absolute effectiveness is investigated to demonstrate interethnic differences in the benefits of 4RIF versus the standard INH treatments. The results demonstrate disparities in the comparative effectiveness across ethnic groups: 9INH was the least effective for all ethnic groups; 6INH significantly improved effectiveness over 9INH among all nonAsian groups; 6H4R significantly improved effectiveness over 6INH for all non-Asian groups; 9H4R significantly improved effectiveness over 6H4R for all non-US born black patient groups. For absolute effectiveness, 6H4R was the most competitive regimen. 6H4R was equally effective as 4RIF for Asian, white and Hispanic patients, assuming a 70% adherence rate on 4RIF. However, 4RIF was dominant over all INHbased regimens when assumed adherence increased to 90%. When accounting for selection in a clinic sample, 9H4R has the greatest overall benefit. Disregarding selection leads to the second-best alternative (i.e., 6H4R) for most ethnic groups. Motivation of comparative effectiveness research Comparative effectiveness research (CER) is designed to improve understanding of individual treatment effects by analyzing subgroup effects and developing clinical prediction rules that improve patient care [18,19] . Pharmacogenomics has been the primary field for CER application. Khoury et al. [20] proposed evaluating the efficacy of pharmacogenomics applications of CER with respect to clinical validity and clinical utility. For the former, they concluded the role of CER for gene-based test applications was to determine whether knowledge of a specific genetic marker better predicts outcomes related to adverse effects as compared with nongenetic predictors. For clinical utility, the value of CER was to assess the added benefits and harms of genetic testing compared with other methods. For LTBI, the promise of improved genetic knowledge centers on predicting the occurrence of side effects from 9INH. Patients of a slow acetylation (SA) phenotype metabolize INH more slowly and
future science group
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
may experience more side effects. Candidate genes for this phenotype include NAT2 [21] and CYP2E1 [22] . The SA phenotype is significantly correlated with the NAT2 polymorphism [23,24] , while the correlation between CYP2E1 and SA phenotype is not as well studied. Recent work has indicated the benefits of a NAT2 genotype guided regimen for the treatment of tuberculosis in an Asian sample [25] . Older work suggested that the SA phenotype is positively associated with the risk of liver injury among patients taking preventive INH [26] . Most side effects to INH occur as elevated liver enzymes, particularly ALT [27] . However, acetylation genotype can also predict other side effects. Orzechowska-Juzwenko et al. [28] and GawronskaSzklarz et al. [29] showed that patients with allergic conditions were more likely to be slow acetylators. Kozhekbaeva et al. [30] found certain NAT2 polymorphisms were risk factors for psoriasis. Behaviors such as smoking and habitual alcohol consumption increase risk further. Thus, certain behaviors exacerbate existing genetic predisposition to side effects. Given that such polymorphisms can affect response to treatment, it is interesting to consider the possibility of testing the genetics of each patient seeking LTBI treatment. Ideally, 4RIF replaces 9INH for patients with a NAT2 polymorphism, thereby reducing the incidence of side effects and improving adherence. Despite minimizing side effects, genetic tests may not be cost effective. These tests are expensive and the patient may gain knowledge of their risk for other diseases, such as cancer [31–33] . The potential for mixed-messaging in patient communication may also grow substantially (Box 1, point 1 & Supplementary Material ; see online at: www.futuremedicine.com/doi/suppl/10.2217/cer.15.3). Such scenarios cannot be ignored given that patients’ motivations to modify their behaviors based on knowledge of their personalized genetic information has been documented [34,35] . We introduce an ethnically diverse clinic population that serves as the basis for our exploration of treatment effectiveness across ethnic groups. We consider ethnicity as a nongenetic predictor of side effects from INH to assess comparative effectiveness. Xie et al. [15] document the ethnic heterogeneity of the relevant NAT2 polymorphism, with the occurrence being greater in non-Asian populations by a magnitude of roughly five. Shenfield [14] and Nerbert [13] noted how Caucasian and black populations have an equal chance of having the SA phenotype, while in Asian populations, the prevalence is only about 10%. Thus, there may be enough phenotypic heterogeneity corresponding to ethnicity to make an effectiveness inquiry into treatment for LTBI a worthwhile investigation.
future science group
Research Article
Comparative effectiveness inquiries 9H4R vs ... 9INH
6INH
6H4R
Figure 1. Comparative effectiveness inquiries. 6H4R: 6 months of isoniazid with 4 months of transitional rifampin; 6INH: 6 months of isoniazid; 9H4R: 9 months of isoniazid with 4 months of transitional rifampin; 9INH: 9 months of isoniazid.
Study population & methods The study sample includes all patients who either started or completed treatment for LTBI in 2010 at the Columbus Public Health TB clinic in Columbus, OH, USA. Thus patients who either started treatment in 2009 but completed in 2010 or started in 2010 and concluded in 2011 were included. The sample consists of ethnically diverse patients who presented with a previous diagnosis of LTBI or a positive test for LTBI. Figure 3 shows the continent of birth for all LTBI patients. Asians and Hispanics (patients born in Latin or South America) accounted for almost a quarter of patients. The data include repeated observations of clinical markers, including ALT, for the duration of treatment, generally 9 months. A complete blood count with differential and hepatic function panel were conducted at baseline, with follow-up liver function tests (LFTs) after the second and fifth months of treatment. Patients were evaluated monthly for adherence to therapy and for possible side effects when medication was dispensed. Therapy was deemed complete if patients attended all scheduled clinic visits and stated they consumed all dispensed medication. Figure 4 gives the treatment time line along with the costs. For all but the first, second and fifth visits, costs consist of a nursing visit and medication. For standard treatment outlined in the figure, expected treatment costs are $240 per patient. However, more (fewer) tests were done at the physician’s discretion when patients were deemed at greater (lower) risk of experiencing adverse effects (i.e., previous illicit drug use, heavy alcohol use, advanced age; however ethnicity was not used as a diagnostic risk factor) or actually experienced Absolute effectiveness inquiries Simulated 4RIF vs ... 9H4R
9INH
6INH
6H4R
Figure 2. Absolute effectiveness inquiries. 4RIF: 4 months of rifampin; 6H4R: 6 months of isoniazid with 4 months of transitional rifampin; 6INH: 6 months of isoniazid; 9H4R: 9 months of isoniazid with 4 months of transitional rifampin; 9INH: 9 months of isoniazid.
www.futuremedicine.com
241
Research Article Fluegge & Roe
Box 1. Additional notes. • Point 1: Having the rapid acetylation phenotype (no NAT2 polymorphism) is deleterious when the exposure is harmful (i.e., environmental tobacco smoke, for example) [36] , yet beneficial for purposes of isoniazid (INH) metabolism. • Point 2: In the USA, the prevalence of latent tuberculosis infection (LTBI) is estimated at slightly over 11 million cases in 2000 (rate of 4.2%), with approximately 20% of that burden consisting of foreign-born cases [37,38] . The prevalence rate of 4.2% represents a 60% decline in LTBI since 1971 [39] ; thus the LTBI prevalence has declined by roughly 20% per decade. To produce an LTBI catchment area statistic for 2010, we multiplied the total population of Franklin County in 2010 by a 3.4% (4.2 × 0.80) LTBI statistic to generate the total number of patients expected to have LTBI, or 36,345 people. • Point 3: Countries included in the rapid acetylation group are Myanmar (Burma), Viet Nam, Bhutan, Bangladesh, Laos, the Philippines, South Korea, North Korea, Malaysia, Taiwan, Cambodia, Hong Kong and Thailand. Not all Asian patients are categorized in the RA group. Countries omitted include India, Iraq and Russia. Ten patients from these countries were included in the slow acetylation group. • Point 4: Genetic susceptibility is the unobserved prevalence of polymorphisms within each ethnic group. We introduce this term as a random intercept with a residual draw occurring at the acetylation phenotype level. A draw is a realization of the residual. Each month one or more residuals are realized and combined with other regressors to form the cost outcome. Different realizations are different ‘draws.’ This term captures unobserved heterogeneity due to genetic factors which are specific to the patient’s ethnic group, but not the individual patient. Genetic sensitivity describes the random influence of dose (in months) of INH medication on side effects. This accounts for atopic patients who experience a genetically induced allergic reaction (rash, nausea, etc.). We introduce this term as a random slope on month of treatment with a residual draw occurring at level-3. This latent variable captures patients with not only a genetic susceptibility, but isolates those patients who are most physically sensitive to INH. Behavioral vulnerability describes behaviors that can result in higher baseline ALT, including obesity, smoking and exercise [40,41] , coffee and caffeine consumption [42] and diabetes [43] . Therefore, we consider patients with higher baseline liver enzymes vulnerable to side effects due to their individual behavior. We verified that patients with a slow acetylation phenotype are not more likely to have higher levels of liver enzymes than rapid acetylation phenotype. We reviewed the literature to locate studies that investigated NAT2 specifically and where baseline liver enzymes were taken to determine if there were differences between acetylation phenotypes. Baseline liver enzymes in each acetylation group were ‘within normal limits’ [16,44–46] . This review informs our assumption that = 0 in the hazard regression. In other words, there is no relevant random effect of phenotype profile in the association between a patient’s baseline ALT measure and incidence of a side effect. Finally, individual heterogeneity captures variation at the patient level not otherwise captured by baseline ALT. We introduce this term as a random intercept with a residual draw occurring at the individual level. • Point 5: The hazard function is modeled with a piecewise linear spline (generalized Gompertz) [47,48] . These estimates are available in the Supplementary Material. • Point 6: Month of treatment is included since month of treatment does not necessarily imply the same days of treatment for each patient. Many patients missed appointments. Given the incidence of side effects may depend on bioaccumulation of the drug, this could affect incidence of side effects. • Point 7: We assume these are negligible, as this is a key benefit of rifampin compared with INH. Liver toxicity from rifampin occurs rarely [49,50] .
side effects warranting extra LFT monitoring. While not advocated by the CDC, additional assessment of liver function has been shown to be beneficial in reducing and preventing asymptomatic liver damage, reducing hospitalization rates and reducing noncompliance with treatment in cases of active tuberculosis [51] . The repeated clinical observations are generally restricted to measuring levels of liver toxicity associated with INH using LFTs. At the TB clinic, liver enzymes were monitored monthly once the patient experienced liver toxicity. Monitoring was discontinued after tests returned to normal. This additional monitoring represents a cost associated with a side effect.
242
J. Comp. Eff. Res. (2015) 4(3)
Figure 5 shows the adherence path for LTBI patients. The top box approximates the catchment area of all LTBI patients in Central Ohio (Box 1, point 2). The clinic sample includes about 2% of those expected to have LTBI. About 55% of patients completed treatment, similar to previous years. African-born patients represented over half (58%) of the nonadherence immediately after the first month of medication as well as claiming the majority of nonadherent cases throughout treatment. Table 1 describes the clinic sample using the 9H4R regimen. Patients who completed treatment were less educated and foreign born. Asian patients were more
future science group
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
likely to complete therapy. Patients with insurance (managed care plans or Medicaid/Medicare) completed treatment more often than patients without insurance. Even though this clinic offers discounted services for low-income patients, insurance coverage and nonadherence were significant correlates. Healthier patients (lower baseline weight, nonsmoker, no history of drug or alcohol use) completed treatment with greater frequency. The last three rows give relevant cost estimates. Thirty four percent of patients had costs associated with side effects. Patients without side effects completed treatment at a mean cost of $236.98, whereas patients with side effects completed treatment at a mean cost that was almost 50% higher (+$349.64). The lead author performed individual record review of all sample patients. Institutional review board approval was granted by the Ohio State University in 2011. Methodology We use a methodology that leverages the comparative effectiveness of 9H4R across ethnic sub-groups to calculate a net monetary benefit (NMB) statistic. The NMB is a monetary value of LTBI treatment adherence, in other words, the willingness to pay for an extra unit of effect net of the additional costs of achieving the effect. Adapting Hoch’s (2002) [52] notation, we calculate net monetary difference (NMD) of 9H4R as:
Research Article
5% 19%
54%
US (89%) 22%
Africa US/Caribbean/ W. Europe Asia Hispanic
Somalia (48%)
Figure 3. Ethnic variation of patients presenting to clinic (n = 552).
K is the mean willingness to pay for a marginal increase in effect (discussed below),
and
are the mean effect and cost, respectively, among the referent Asian population (i = 1), and and are the mean effects and costs among each of the other four ethnic groups (i = 2, 3, 4 and 5). and with i = 2, 3, 4 and 5 are the mean differences in the costs and health effects between ethnic groups. For example, = gives mean NMD between Asian and white patients. Subscripts 3, 4 and 5 refer to US born blacks, Hispanics and Africans, respectively. Mean health effects ( ) are between zero and one and represent the proportion of an ethnic group com-
LTBI routine treatment timeline and costs, Columbus Public Health, Columbus, OH, 2010 Month 3 medication ($1.05), NV ($22.88) Month 4 medication ($1.05), NV ($22.88) Month 5 medication ($1.05), Second routine LFT ($42.13)
Pretreatment clinic visits (TST, QFT, CXR, HIV, CBC, AFB – costs excluded)
-1
0
1
2
3
Month 2 medication ($1.05), First routine LFT ($42.13) Medical exam ($37)/begin therapy month 1 (INH = $1.05)
4
5
6
Month 8 medication ($1.05), NV ($22.88)
7
8
9
Month 7 medication ($1.05), Completed, no return visit NV ($22.88) Month 9 medication ($1.05), Month 6 medication ($1.05), NV ($22.88) NV ($22.88)
Figure 4. Treatment time line. AFB: Acid-fast bacilli; CBC: Complete blood count; CXR: Chest x-ray; INH: Isoniazid; LTBI: Latent tuberculosis infection; NV: Nursing visit; LFT: Liver function test; QFT: QuantiFERON; TST: Tuberculin skin test.
future science group
www.futuremedicine.com
243
Research Article Fluegge & Roe
2010 LTBI catchment population: 1,068,978 × 3.4% = 36,345
36 tuberculin skin tests, 170 quantiferon tests
552 patients diagnosed with LTBI (2% of catchment population)
Ethnic composition: 7 Asian, 10 white, 22 US born black, 2 Hispanic, 55 African blacks
96 (17%) patients dropped out after receiving month 1 INH
Ethnic composition: 78 Asian, 58 white, 96 US born black, 27 Hispanic, 293 African black
154 (28%) patients dropped out before completing
Ethnic composition: 13 Asian, 18 white, 31 US born black, 7 Hispanic, 85 African black
Mean cost of side effects – rapid acetylators (n = 30): $349.45 (SD = 102.80, range: 180.09–665.39)
302 (55%) patients completed LTBI therapy
Mean cost of side effects – slow acetylators (n = 112): $348.30 (SD = 107.90, range: 181.03–800.25)
Ethnic composition (completed): 9 Asian, 1 white, 5 US born black, 4 Hispanic, 13 African black
42 patients switched to rifampin; 32 patients completed on rifampin
Ethnic composition (all): 11 Asian, 3 white, 8 US born black, 5 Hispanic, 15 African black
Figure 5. Adherence data for 2010 latent tuberculosis infection clinic population. INH: Isoniazid; LTBI: Latent tuberculosis infection; SD: Standard deviation.
244
J. Comp. Eff. Res. (2015) 4(3)
future science group
future science group
0.59 0.77 0.57
96 94 96 96 96
Male*
Education***
Foreign born*
African
US black
95
No insurance***
15
Costs of prevention ($) | SE = 0
42.28
88.51
49.14 0, 104.34
0, 253.43
0, 113.92
4, 73
105.7, 367.3
0, 1
0, 1
0, 1
0, 1
0, 31
0, 1
0, 1
0, 1
0, 1
0.45, 19.2
0, 3300
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 16
0, 1
18, 81
48.40
88.58
7.65
13.25
46.61
0.29
0.48
0.37
0.47
6.88
0.48
0.26
0.33
0.36
4.45
829.91
0.31
0.14
0.26
0.42
0.50
0.42
4.65
0.49
14.38
SD
7
46
101
154
154
154
154
154
154
154
154
154
154
154
153
133
154
154
154
154
154
154
154
154
154
n
106.68
232.45
128.34
23.71
165.77
0.12
0.33
0.14
0.34
5.38
0.56
0.06
0.12
0.26
8.36
817.00
0.12
0.05
0.08
0.20
0.55
0.75
10.73
0.45
38.85
Mean
0, 278.66
90, 653.3
46.81, 329.39
5, 133
83, 284
0, 1
0, 1
0, 1
0, 1
0, 29
0, 1
0, 1
0, 1
0, 1
0.3, 23.9
0, 8333
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 18
0, 1
20, 73
Min, max
1 ≤ months ≤ 9 (n = 154)
94.16
120.8
24.31
16.64
36.76
0.33
0.47
0.34
0.48
6.68
0.50
0.25
0.32
0.44
4.24
1109
0.32
0.32
0.28
0.40
0.50
0.44
4.33
0.50
12.14
SD
8
136
158
302
302
301
301
301
301
302
302
302
302
302
302
279
302
302
302
302
302
302
293
302
302
n
267.07
349.64
237.0
25.32
158.1
0.10
0.25
0.08
0.23
3.72
0.47
0.07
0.19
0.29
9.10
730.3
0.10
0.06
0.19
0.15
0.51
0.84
9.42
0.52
39.74
0,380.57
180, 800
143.6, 396.7
5, 201
83, 353
0, 1
0, 1
0, 1
0, 1
0, 40
0, 1
0, 1
0, 1
0, 1
0.19, 18.6
0, 4800
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 16
0, 1
18, 88
Mix, max
Completed (n = 302) Mean
121
108.5
51.60
19.85
36.71
0.30
0.43
0.28
0.42
6.52
0.50
0.25
0.39
0.46
3.37
771.1
0.30
0.24
0.39
0.35
0.50
0.37
5.31
0.50
14.82
SD
ALT is measured to see if the liver is damaged or diseased. When the liver is damaged or diseased, it releases ALT into the bloodstream, which makes ALT levels go up. ALT is more sensitive to isoniazid than AST. χ2 (binary) and one-way ANOVA (continuous): *p < 0.10. **p < 0.05. ***p < 0.01. LTBI: Latent tuberculosis infection.
6
Costs of SE ($) | SE = 1
22.02
95 75
Mean ALT
Cost of treatment ($)
171.84
96
Intake weight (lbs)***
0.35 0.09
96 96
Prev LTBI patient
0.17
0.31
4.44
0.64
0.07
0.13
0.15
0.58
Alcohol use*
96
95
Illicit drug use**
95
Medicaid***
Private ins. 96
95
Managed care**
96
95
Tobacco user**
698.68
80
Mo. Income
Clinic distance (mi)
Years in US, foreign born**
0.10
96
US white
0.02
0.07
96 96
Asian***
Hispanic
0.23
10.91
39.31
96
Age
Min, max
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
Aim: Nine months of isoniazid (9INH) is the gold standard for treatment of latent tuberculosis infection (LTBI). This paper compares the effectiveness of 9 months of isoniazid with 4 months of transitional rifampin (9H4R) to alternative therapies, including 9INH, 6 months of isoniazid (6INH) and 6 months of isoniazid with 4 months of transitional rifampin (6H4R), for treatment of LTBI. Materials & methods: Using an ethnically diverse clinic sample of 552 patients given treatment for LTBI with 9H4R, we use multilevel selection models to examine the adjusted comparative effectiveness of the regimens among ethnic groups that feature distinct genetic predispositions to side effects on INH. For unadjusted/absolute effectiveness, we simulated cost– effectiveness ratios for 4 months of rifampin (4RIF) and compared with bootstrapped confidence intervals for the alternative therapies. Results: There are variations in the comparative effectiveness across ethnic groups, with the most notable differences for 9H4R. For unadjusted/absolute effectiveness, 4RIF presents the greatest net benefit for US born black and African patients. For all other ethnic groups, 6H4R was the most effective. Conclusion: Patient ethnicity affects tolerance to INH. 9H4R was the most effective LTBI treatment for all ethnicities. However, this result heavily depends on whether adjustments are made for self-selection.
Kyle R Fluegge*,1,2,3,4 & Brian E Roe1 Department of Agricultural, Environmental & Development Economics, Ohio State University, 2120 Fyffe Road, Columbus, OH 43210, USA 2 Division of Epidemiology, College of Public Health, Ohio State University, 2120 Fyffe Road, Columbus, OH 43210, USA 3 Institute for Health & Environmental Research, Columbus, OH 43220, USA 4 Department of Epidemiology & Biostatistics, Case Western Reserve University, 2103 Cornell Road, Cleveland, OH 44106, USA *Author for correspondence: [email protected] 1
Keywords: comparative effectiveness • cost–effectiveness • ethnicity • isoniazid • latent tuberculosis infection • Ohio • rifampin
Tuberculosis (TB) remains a major global health concern. In 2010, there were 8.8 million worldwide incident cases, from which more than a million individuals die annually [1] . Furthermore, the Centers for Disease Control and Prevention (CDC) estimates that a third of the world’s population is infected with Mycobacterium tuberculosis [2] . This constitutes a large reservoir of patients with latent tuberculosis infection (LTBI), some of whom will develop active disease if not treated [3] . Two LTBI treatment regimens have received the most attention: 9 months of daily isoniazid (9INH) and 4 months of daily rifampin (4RIF). 9INH is less expensive. However, the 4RIF treatment is shorter and causes fewer side effects, which often results in greater rates of respondent completion for 4RIF. Side effects may be, in part, geneti-
10.2217/CER.15.3 © 2015 Future Medicine Ltd
cally determined, which gives rise to questions of comparative effectiveness for patient groups with substantial genetic variation. In this paper, we study a clinically derived LTBI treatment regimen: 9INH with transitional 4RIF (9H4R) if INH is not well tolerated. As a result, we are able to more fully assess outcomes for patients who would have otherwise had 9INH therapy terminated as well as comparing outcomes to several alternative treatment definitions including 9INH alone, 6 months of isoniazid (6INH) or 6INH with transitional 4RIF (6H4R). The questions of relevance in this paper are does the relative effectiveness of 9H4R, 9INH, 6INH or 6H4R differ across patient ethnicity in a predictable fashion and is any one or more of these regimens more or less cost effective than a simulated regimen of 4RIF alone?
J. Comp. Eff. Res. (2015) 4(3), 239–257
part of
ISSN 2042-6305
239
Research Article Fluegge & Roe Multiple studies consider the relative cost–effectiveness (CE) of 4RIF, 9INH and 6INH. Most consist of CE analyses done in the context of randomized-controlled trials (RCTs) [4–8] , whereas others use retro spective clinical data [9] . RCTs are widely considered the superior methodology to assess treatment outcomes and their costs. The process of randomization in RCTs is thought to prevent bias that occurs with participant self-selection. In other words, randomization of patients to treatment removes patient heterogeneity that may be correlated with treatment outcomes. However, randomization may be unable to eliminate bias due to patients who are unwilling to participate (i.e., the unenrolled). In this case, researchers are faced with potential systematic differences between the trial and general patient population. Bias may also result from unmeasured and unpredicted effects of treatment on the outcomes of interest, choice of randomizing variables that are correlated with treatment benefits [10] or complete omission of randomization when it is needed. In this paper we describe a single-arm comparative effectiveness methodology [11,12] with ethnic subgroup analyses to test the effectiveness of 9H4R, 9INH, 6INH and 6H4R, and examine the absolute effectiveness of these regimens compared with a simulation of care with only 4RIF. These inquiries are shown in Figures 1 & 2. By a single-arm, we refer to a design wherein all patients are started on the same medication (INH), but may have switched to rifampin as a result of side effects on INH (the 9H4R and 6H4R definitions). We use a sample of 552 clinic patients with medically confirmed diagnoses of LTBI that began INH. Adherence is determined by the treatment definitions (i.e., 9H4R, 9INH, 6INH and 6H4R). To assess the comparative effectiveness, we use a series of multilevel regression models to analyze patient selection into treatment, patient attrition from treatment and accumulated costs of treatment, where the multiple levels involve treatment months, individual patients and a grouping variable based on patient ethnicity. All models control for patient ethnicity because different ethnic groups acetylate (metabolize) isoniazid differently [13,14] . Asian patients, who are expected to best tolerate 9INH [15,16] , serve as the reference group. To analyze absolute effectiveness, we compute bootstrapped confidence intervals for the CE ratios of 9H4R, 9INH, 6INH and 6H4R for each ethnic group. Simulated CE ratios of 4RIF are compared with these confidence intervals to determine if 4RIF improves treatment effectiveness more than the various INH-based therapies. The contributions of this work inform the literature in several ways. First, we are the first to investigate the
240
J. Comp. Eff. Res. (2015) 4(3)
potential of transitional LTBI therapy (i.e., 9H4R, 6H4R), which has been suggested as a potentially effective modified protocol for patients experiencing side effects on INH [17] . Second, we incorporate selection models into the comparative effectiveness analyses to address the dual nature of selection; that is, selection into treatment and attrition from treatment, which are largely disregarded in RCTs. Third, we use a comparative effectiveness modeling framework with random effects (intercepts and slopes) to specifically model net benefit differences (i.e., effectiveness) between ethnic groups. Last, absolute effectiveness is investigated to demonstrate interethnic differences in the benefits of 4RIF versus the standard INH treatments. The results demonstrate disparities in the comparative effectiveness across ethnic groups: 9INH was the least effective for all ethnic groups; 6INH significantly improved effectiveness over 9INH among all nonAsian groups; 6H4R significantly improved effectiveness over 6INH for all non-Asian groups; 9H4R significantly improved effectiveness over 6H4R for all non-US born black patient groups. For absolute effectiveness, 6H4R was the most competitive regimen. 6H4R was equally effective as 4RIF for Asian, white and Hispanic patients, assuming a 70% adherence rate on 4RIF. However, 4RIF was dominant over all INHbased regimens when assumed adherence increased to 90%. When accounting for selection in a clinic sample, 9H4R has the greatest overall benefit. Disregarding selection leads to the second-best alternative (i.e., 6H4R) for most ethnic groups. Motivation of comparative effectiveness research Comparative effectiveness research (CER) is designed to improve understanding of individual treatment effects by analyzing subgroup effects and developing clinical prediction rules that improve patient care [18,19] . Pharmacogenomics has been the primary field for CER application. Khoury et al. [20] proposed evaluating the efficacy of pharmacogenomics applications of CER with respect to clinical validity and clinical utility. For the former, they concluded the role of CER for gene-based test applications was to determine whether knowledge of a specific genetic marker better predicts outcomes related to adverse effects as compared with nongenetic predictors. For clinical utility, the value of CER was to assess the added benefits and harms of genetic testing compared with other methods. For LTBI, the promise of improved genetic knowledge centers on predicting the occurrence of side effects from 9INH. Patients of a slow acetylation (SA) phenotype metabolize INH more slowly and
future science group
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
may experience more side effects. Candidate genes for this phenotype include NAT2 [21] and CYP2E1 [22] . The SA phenotype is significantly correlated with the NAT2 polymorphism [23,24] , while the correlation between CYP2E1 and SA phenotype is not as well studied. Recent work has indicated the benefits of a NAT2 genotype guided regimen for the treatment of tuberculosis in an Asian sample [25] . Older work suggested that the SA phenotype is positively associated with the risk of liver injury among patients taking preventive INH [26] . Most side effects to INH occur as elevated liver enzymes, particularly ALT [27] . However, acetylation genotype can also predict other side effects. Orzechowska-Juzwenko et al. [28] and GawronskaSzklarz et al. [29] showed that patients with allergic conditions were more likely to be slow acetylators. Kozhekbaeva et al. [30] found certain NAT2 polymorphisms were risk factors for psoriasis. Behaviors such as smoking and habitual alcohol consumption increase risk further. Thus, certain behaviors exacerbate existing genetic predisposition to side effects. Given that such polymorphisms can affect response to treatment, it is interesting to consider the possibility of testing the genetics of each patient seeking LTBI treatment. Ideally, 4RIF replaces 9INH for patients with a NAT2 polymorphism, thereby reducing the incidence of side effects and improving adherence. Despite minimizing side effects, genetic tests may not be cost effective. These tests are expensive and the patient may gain knowledge of their risk for other diseases, such as cancer [31–33] . The potential for mixed-messaging in patient communication may also grow substantially (Box 1, point 1 & Supplementary Material ; see online at: www.futuremedicine.com/doi/suppl/10.2217/cer.15.3). Such scenarios cannot be ignored given that patients’ motivations to modify their behaviors based on knowledge of their personalized genetic information has been documented [34,35] . We introduce an ethnically diverse clinic population that serves as the basis for our exploration of treatment effectiveness across ethnic groups. We consider ethnicity as a nongenetic predictor of side effects from INH to assess comparative effectiveness. Xie et al. [15] document the ethnic heterogeneity of the relevant NAT2 polymorphism, with the occurrence being greater in non-Asian populations by a magnitude of roughly five. Shenfield [14] and Nerbert [13] noted how Caucasian and black populations have an equal chance of having the SA phenotype, while in Asian populations, the prevalence is only about 10%. Thus, there may be enough phenotypic heterogeneity corresponding to ethnicity to make an effectiveness inquiry into treatment for LTBI a worthwhile investigation.
future science group
Research Article
Comparative effectiveness inquiries 9H4R vs ... 9INH
6INH
6H4R
Figure 1. Comparative effectiveness inquiries. 6H4R: 6 months of isoniazid with 4 months of transitional rifampin; 6INH: 6 months of isoniazid; 9H4R: 9 months of isoniazid with 4 months of transitional rifampin; 9INH: 9 months of isoniazid.
Study population & methods The study sample includes all patients who either started or completed treatment for LTBI in 2010 at the Columbus Public Health TB clinic in Columbus, OH, USA. Thus patients who either started treatment in 2009 but completed in 2010 or started in 2010 and concluded in 2011 were included. The sample consists of ethnically diverse patients who presented with a previous diagnosis of LTBI or a positive test for LTBI. Figure 3 shows the continent of birth for all LTBI patients. Asians and Hispanics (patients born in Latin or South America) accounted for almost a quarter of patients. The data include repeated observations of clinical markers, including ALT, for the duration of treatment, generally 9 months. A complete blood count with differential and hepatic function panel were conducted at baseline, with follow-up liver function tests (LFTs) after the second and fifth months of treatment. Patients were evaluated monthly for adherence to therapy and for possible side effects when medication was dispensed. Therapy was deemed complete if patients attended all scheduled clinic visits and stated they consumed all dispensed medication. Figure 4 gives the treatment time line along with the costs. For all but the first, second and fifth visits, costs consist of a nursing visit and medication. For standard treatment outlined in the figure, expected treatment costs are $240 per patient. However, more (fewer) tests were done at the physician’s discretion when patients were deemed at greater (lower) risk of experiencing adverse effects (i.e., previous illicit drug use, heavy alcohol use, advanced age; however ethnicity was not used as a diagnostic risk factor) or actually experienced Absolute effectiveness inquiries Simulated 4RIF vs ... 9H4R
9INH
6INH
6H4R
Figure 2. Absolute effectiveness inquiries. 4RIF: 4 months of rifampin; 6H4R: 6 months of isoniazid with 4 months of transitional rifampin; 6INH: 6 months of isoniazid; 9H4R: 9 months of isoniazid with 4 months of transitional rifampin; 9INH: 9 months of isoniazid.
www.futuremedicine.com
241
Research Article Fluegge & Roe
Box 1. Additional notes. • Point 1: Having the rapid acetylation phenotype (no NAT2 polymorphism) is deleterious when the exposure is harmful (i.e., environmental tobacco smoke, for example) [36] , yet beneficial for purposes of isoniazid (INH) metabolism. • Point 2: In the USA, the prevalence of latent tuberculosis infection (LTBI) is estimated at slightly over 11 million cases in 2000 (rate of 4.2%), with approximately 20% of that burden consisting of foreign-born cases [37,38] . The prevalence rate of 4.2% represents a 60% decline in LTBI since 1971 [39] ; thus the LTBI prevalence has declined by roughly 20% per decade. To produce an LTBI catchment area statistic for 2010, we multiplied the total population of Franklin County in 2010 by a 3.4% (4.2 × 0.80) LTBI statistic to generate the total number of patients expected to have LTBI, or 36,345 people. • Point 3: Countries included in the rapid acetylation group are Myanmar (Burma), Viet Nam, Bhutan, Bangladesh, Laos, the Philippines, South Korea, North Korea, Malaysia, Taiwan, Cambodia, Hong Kong and Thailand. Not all Asian patients are categorized in the RA group. Countries omitted include India, Iraq and Russia. Ten patients from these countries were included in the slow acetylation group. • Point 4: Genetic susceptibility is the unobserved prevalence of polymorphisms within each ethnic group. We introduce this term as a random intercept with a residual draw occurring at the acetylation phenotype level. A draw is a realization of the residual. Each month one or more residuals are realized and combined with other regressors to form the cost outcome. Different realizations are different ‘draws.’ This term captures unobserved heterogeneity due to genetic factors which are specific to the patient’s ethnic group, but not the individual patient. Genetic sensitivity describes the random influence of dose (in months) of INH medication on side effects. This accounts for atopic patients who experience a genetically induced allergic reaction (rash, nausea, etc.). We introduce this term as a random slope on month of treatment with a residual draw occurring at level-3. This latent variable captures patients with not only a genetic susceptibility, but isolates those patients who are most physically sensitive to INH. Behavioral vulnerability describes behaviors that can result in higher baseline ALT, including obesity, smoking and exercise [40,41] , coffee and caffeine consumption [42] and diabetes [43] . Therefore, we consider patients with higher baseline liver enzymes vulnerable to side effects due to their individual behavior. We verified that patients with a slow acetylation phenotype are not more likely to have higher levels of liver enzymes than rapid acetylation phenotype. We reviewed the literature to locate studies that investigated NAT2 specifically and where baseline liver enzymes were taken to determine if there were differences between acetylation phenotypes. Baseline liver enzymes in each acetylation group were ‘within normal limits’ [16,44–46] . This review informs our assumption that = 0 in the hazard regression. In other words, there is no relevant random effect of phenotype profile in the association between a patient’s baseline ALT measure and incidence of a side effect. Finally, individual heterogeneity captures variation at the patient level not otherwise captured by baseline ALT. We introduce this term as a random intercept with a residual draw occurring at the individual level. • Point 5: The hazard function is modeled with a piecewise linear spline (generalized Gompertz) [47,48] . These estimates are available in the Supplementary Material. • Point 6: Month of treatment is included since month of treatment does not necessarily imply the same days of treatment for each patient. Many patients missed appointments. Given the incidence of side effects may depend on bioaccumulation of the drug, this could affect incidence of side effects. • Point 7: We assume these are negligible, as this is a key benefit of rifampin compared with INH. Liver toxicity from rifampin occurs rarely [49,50] .
side effects warranting extra LFT monitoring. While not advocated by the CDC, additional assessment of liver function has been shown to be beneficial in reducing and preventing asymptomatic liver damage, reducing hospitalization rates and reducing noncompliance with treatment in cases of active tuberculosis [51] . The repeated clinical observations are generally restricted to measuring levels of liver toxicity associated with INH using LFTs. At the TB clinic, liver enzymes were monitored monthly once the patient experienced liver toxicity. Monitoring was discontinued after tests returned to normal. This additional monitoring represents a cost associated with a side effect.
242
J. Comp. Eff. Res. (2015) 4(3)
Figure 5 shows the adherence path for LTBI patients. The top box approximates the catchment area of all LTBI patients in Central Ohio (Box 1, point 2). The clinic sample includes about 2% of those expected to have LTBI. About 55% of patients completed treatment, similar to previous years. African-born patients represented over half (58%) of the nonadherence immediately after the first month of medication as well as claiming the majority of nonadherent cases throughout treatment. Table 1 describes the clinic sample using the 9H4R regimen. Patients who completed treatment were less educated and foreign born. Asian patients were more
future science group
A comparative effectiveness analysis of treatment for latent tuberculosis infection using multilevel selection models
likely to complete therapy. Patients with insurance (managed care plans or Medicaid/Medicare) completed treatment more often than patients without insurance. Even though this clinic offers discounted services for low-income patients, insurance coverage and nonadherence were significant correlates. Healthier patients (lower baseline weight, nonsmoker, no history of drug or alcohol use) completed treatment with greater frequency. The last three rows give relevant cost estimates. Thirty four percent of patients had costs associated with side effects. Patients without side effects completed treatment at a mean cost of $236.98, whereas patients with side effects completed treatment at a mean cost that was almost 50% higher (+$349.64). The lead author performed individual record review of all sample patients. Institutional review board approval was granted by the Ohio State University in 2011. Methodology We use a methodology that leverages the comparative effectiveness of 9H4R across ethnic sub-groups to calculate a net monetary benefit (NMB) statistic. The NMB is a monetary value of LTBI treatment adherence, in other words, the willingness to pay for an extra unit of effect net of the additional costs of achieving the effect. Adapting Hoch’s (2002) [52] notation, we calculate net monetary difference (NMD) of 9H4R as:
Research Article
5% 19%
54%
US (89%) 22%
Africa US/Caribbean/ W. Europe Asia Hispanic
Somalia (48%)
Figure 3. Ethnic variation of patients presenting to clinic (n = 552).
K is the mean willingness to pay for a marginal increase in effect (discussed below),
and
are the mean effect and cost, respectively, among the referent Asian population (i = 1), and and are the mean effects and costs among each of the other four ethnic groups (i = 2, 3, 4 and 5). and with i = 2, 3, 4 and 5 are the mean differences in the costs and health effects between ethnic groups. For example, = gives mean NMD between Asian and white patients. Subscripts 3, 4 and 5 refer to US born blacks, Hispanics and Africans, respectively. Mean health effects ( ) are between zero and one and represent the proportion of an ethnic group com-
LTBI routine treatment timeline and costs, Columbus Public Health, Columbus, OH, 2010 Month 3 medication ($1.05), NV ($22.88) Month 4 medication ($1.05), NV ($22.88) Month 5 medication ($1.05), Second routine LFT ($42.13)
Pretreatment clinic visits (TST, QFT, CXR, HIV, CBC, AFB – costs excluded)
-1
0
1
2
3
Month 2 medication ($1.05), First routine LFT ($42.13) Medical exam ($37)/begin therapy month 1 (INH = $1.05)
4
5
6
Month 8 medication ($1.05), NV ($22.88)
7
8
9
Month 7 medication ($1.05), Completed, no return visit NV ($22.88) Month 9 medication ($1.05), Month 6 medication ($1.05), NV ($22.88) NV ($22.88)
Figure 4. Treatment time line. AFB: Acid-fast bacilli; CBC: Complete blood count; CXR: Chest x-ray; INH: Isoniazid; LTBI: Latent tuberculosis infection; NV: Nursing visit; LFT: Liver function test; QFT: QuantiFERON; TST: Tuberculin skin test.
future science group
www.futuremedicine.com
243
Research Article Fluegge & Roe
2010 LTBI catchment population: 1,068,978 × 3.4% = 36,345
36 tuberculin skin tests, 170 quantiferon tests
552 patients diagnosed with LTBI (2% of catchment population)
Ethnic composition: 7 Asian, 10 white, 22 US born black, 2 Hispanic, 55 African blacks
96 (17%) patients dropped out after receiving month 1 INH
Ethnic composition: 78 Asian, 58 white, 96 US born black, 27 Hispanic, 293 African black
154 (28%) patients dropped out before completing
Ethnic composition: 13 Asian, 18 white, 31 US born black, 7 Hispanic, 85 African black
Mean cost of side effects – rapid acetylators (n = 30): $349.45 (SD = 102.80, range: 180.09–665.39)
302 (55%) patients completed LTBI therapy
Mean cost of side effects – slow acetylators (n = 112): $348.30 (SD = 107.90, range: 181.03–800.25)
Ethnic composition (completed): 9 Asian, 1 white, 5 US born black, 4 Hispanic, 13 African black
42 patients switched to rifampin; 32 patients completed on rifampin
Ethnic composition (all): 11 Asian, 3 white, 8 US born black, 5 Hispanic, 15 African black
Figure 5. Adherence data for 2010 latent tuberculosis infection clinic population. INH: Isoniazid; LTBI: Latent tuberculosis infection; SD: Standard deviation.
244
J. Comp. Eff. Res. (2015) 4(3)
future science group
future science group
0.59 0.77 0.57
96 94 96 96 96
Male*
Education***
Foreign born*
African
US black
95
No insurance***
15
Costs of prevention ($) | SE = 0
42.28
88.51
49.14 0, 104.34
0, 253.43
0, 113.92
4, 73
105.7, 367.3
0, 1
0, 1
0, 1
0, 1
0, 31
0, 1
0, 1
0, 1
0, 1
0.45, 19.2
0, 3300
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 16
0, 1
18, 81
48.40
88.58
7.65
13.25
46.61
0.29
0.48
0.37
0.47
6.88
0.48
0.26
0.33
0.36
4.45
829.91
0.31
0.14
0.26
0.42
0.50
0.42
4.65
0.49
14.38
SD
7
46
101
154
154
154
154
154
154
154
154
154
154
154
153
133
154
154
154
154
154
154
154
154
154
n
106.68
232.45
128.34
23.71
165.77
0.12
0.33
0.14
0.34
5.38
0.56
0.06
0.12
0.26
8.36
817.00
0.12
0.05
0.08
0.20
0.55
0.75
10.73
0.45
38.85
Mean
0, 278.66
90, 653.3
46.81, 329.39
5, 133
83, 284
0, 1
0, 1
0, 1
0, 1
0, 29
0, 1
0, 1
0, 1
0, 1
0.3, 23.9
0, 8333
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 18
0, 1
20, 73
Min, max
1 ≤ months ≤ 9 (n = 154)
94.16
120.8
24.31
16.64
36.76
0.33
0.47
0.34
0.48
6.68
0.50
0.25
0.32
0.44
4.24
1109
0.32
0.32
0.28
0.40
0.50
0.44
4.33
0.50
12.14
SD
8
136
158
302
302
301
301
301
301
302
302
302
302
302
302
279
302
302
302
302
302
302
293
302
302
n
267.07
349.64
237.0
25.32
158.1
0.10
0.25
0.08
0.23
3.72
0.47
0.07
0.19
0.29
9.10
730.3
0.10
0.06
0.19
0.15
0.51
0.84
9.42
0.52
39.74
0,380.57
180, 800
143.6, 396.7
5, 201
83, 353
0, 1
0, 1
0, 1
0, 1
0, 40
0, 1
0, 1
0, 1
0, 1
0.19, 18.6
0, 4800
0, 1
0, 1
0, 1
0, 1
0, 1
0, 1
0, 16
0, 1
18, 88
Mix, max
Completed (n = 302) Mean
121
108.5
51.60
19.85
36.71
0.30
0.43
0.28
0.42
6.52
0.50
0.25
0.39
0.46
3.37
771.1
0.30
0.24
0.39
0.35
0.50
0.37
5.31
0.50
14.82
SD
ALT is measured to see if the liver is damaged or diseased. When the liver is damaged or diseased, it releases ALT into the bloodstream, which makes ALT levels go up. ALT is more sensitive to isoniazid than AST. χ2 (binary) and one-way ANOVA (continuous): *p < 0.10. **p < 0.05. ***p < 0.01. LTBI: Latent tuberculosis infection.
6
Costs of SE ($) | SE = 1
22.02
95 75
Mean ALT
Cost of treatment ($)
171.84
96
Intake weight (lbs)***
0.35 0.09
96 96
Prev LTBI patient
0.17
0.31
4.44
0.64
0.07
0.13
0.15
0.58
Alcohol use*
96
95
Illicit drug use**
95
Medicaid***
Private ins. 96
95
Managed care**
96
95
Tobacco user**
698.68
80
Mo. Income
Clinic distance (mi)
Years in US, foreign born**
0.10
96
US white
0.02
0.07
96 96
Asian***
Hispanic
0.23
10.91
39.31
96
Age
Min, max
