http://orcid.org/0000-0001-8223-5716
Figures
Abstract
Latent tuberculosis infection (LTBI) treatment in persons at increased risk of disease progression is a key strategy with the strong potential to increase rate of tuberculosis (TB) decline in the United States. However, LTBI treatment in homeless persons, a population at high-risk of active TB disease, is usually associated with poor adherence. We describe the impact of using directly observed treatment (DOT) versus self-administered treatments (SAT) as an adherence-improving intervention to administer four months of daily rifampin regimen for LTBI treatment among homeless adults in Atlanta. Retrospective analysis of clinical care data on 274 homeless persons who initiated daily rifampin treatment for LTBI treatment at a county health department between January 2014 and December 2016 was performed. To reduce bias from non-random assignment of treatment, an inverse probability of treatment weighted (IPTW) logistic regression model was used to assess the effect of treatment type on treatment completion. Subgroup analyses were performed to assess heterogeneity of treatment effect on LTBI completion. Of 274 LTBI treatment initiators, 177 (65%) completed treatment [DOT 118/181 (65%), SAT 59/93 (63%)]. In the fully adjusted and weighted analysis, the odds of completing LTBI treatment on DOT was 40% higher than the odds of completing treatment by SAT [adjusted odds ratio (95% CI), aOR: 1.40 (1.07, 1.82), p = 0.014]. The unstable nature of homeless persons’ lifestyle makes LTBI treatment difficult for many reasons. Our study lends support to the use of DOT to improve LTBI treatment completion among subgroups of homeless persons on treatment with daily rifampin.
Citation: Onwubiko U, Wall K, Sales R-M, Holland DP (2019) Using Directly Observed Therapy (DOT) for latent tuberculosis treatment – A hit or a miss? A propensity score analysis of treatment completion among 274 homeless adults in Fulton County, GA. PLoS ONE 14(6): e0218373. https://doi.org/10.1371/journal.pone.0218373
Editor: Amy Davidow, UMDNJ-New Jersey Medical School, UNITED STATES
Received: November 28, 2018; Accepted: June 2, 2019; Published: June 21, 2019
Copyright: © 2019 Onwubiko et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: Author RMS, formerly a staff of the Georgia Department of Public Health during the study inception, data collection, analysis and initial submission of this paper, is currently affiliated with the Tuberculosis program at Texas Department of Health and Human Services, Austin, Texas. This agency did not have any role in the study design, data collection and analysis, decision to publish or preparation of the manuscript. The specific role of this co-author is described in the author contributions section.
Competing interests: Author RMS, formerly a staff of the Georgia Department of Public Health at the time of the study inception, analysis and submission of this paper, is currently affiliated with Texas Department of Health and Human Services, Austin, Texas. The co-author’s affiliation with this agency does not alter our adherence to all PLOS ONE policies on sharing data and materials.
Introduction
Tuberculosis (TB) is one of the top ten leading causes of death globally.[1] In 2015, about 1.4 million deaths were attributed to TB, making it the global leading cause of death from an infectious disease, ahead of HIV/AIDS.[2] Using strategic disease surveillance and prioritized TB control methods, the current incidence of TB in the United States (US) has been driven to the lowest in recorded history. [3] The role that implementation of the top three priorities of TB control (early identification of active TB cases, prompt initiation of treatment with recommended regimen and the identification with treatment of exposed contacts) has played in driving the decline in active TB in the US has been well documented.[3,4] Epidemiologic modeling, however, has shown that achieving the goal of TB elimination well before the end of this century in all at-risk populations in the US will require a stronger focus on identification and treatment of latent TB infection (LTBI) among persons at high-risk of TB disease.[3–5] These high-risk groups include HIV-infected persons, immigrants and refugees from countries with high TB burden, alcohol and drug abusers and persons dwelling in congregate settings such as incarceration centers and homeless housing facilities.[6,7] With four-fifths (80%) of all active TB disease in the US attributed to reactivation of latent TB infections (LTBI) rather than primary TB infections[8], the effective and prompt treatment of LTBI in these high risk groups remains the key strategy for eliminating TB in the US.
Within the last decade, the field of LTBI therapeutics has seen several innovative advances and research. Newer treatment regimens with shorter durations have been introduced and extensively studied for both efficacy and non-inferiority in comparison to the well-known nine months of daily isoniazid. These new regimens include four months of rifampin and three months of Isoniazid and Rifapentine.[9–13] Despite these advances, LTBI treatment adherence and completion in non-clinical trials setting, particularly among homeless persons, has remained poor.[11,14,15] A recent systematic review of 83 studies on LTBI treatment by Sandgren et al found that proportions of persons completing LTBI treatment among homeless persons was consistently lower (23–71%) than proportions completing treatment in other population subgroups; general population (39–96%), case contacts (48–82%), HIV positive clients (55–95%), immigrants (7–86%) and inmates (4–100%).[16]
While poor adherence to prescribed medical intervention in any population is a major hurdle in achieving desired healthcare outcomes, it is especially problematic when the health condition is a chronic illness that lacks obvious symptoms, presenting no discernable disruption to one’s quality of life.[17] Poor adherence to prescribed LTBI treatment is an important obstacle among homeless persons. This is because adherence to LTBI treatment is the key link between a latent infection and achieving the desired reduction in the risk of progression to active TB. Several adherence-improving interventions have been proposed and explored in the past. One of those explored in relation to TB disease management is directly observed therapy (DOT).[12,15,18–20] Its use in active TB management is well documented. However, DOT’s impact on LTBI treatment among homeless adults on daily rifampin therapy is unclear. We aim to fill that gap by assessing and quantifying the impact of using DOT for administration of four months daily rifampin treatment for LTBI among homeless persons.
Methods
Study design, data source and ethics statement
This was a retrospective observational cohort study. We abstracted demographic and routine clinical data of homeless persons who received TB screening, evaluation and LTBI treatment using 4 months of rifampin with the county health department between January 2014 and December 2016. Ethical approval was obtained from the Institutional Review Board (IRB) of Georgia Department of Public Health (Project #: 170405). All data used were fully anonymized prior to analysis and IRB waived the requirement for informed consent.
Background on study sample
An outbreak of isoniazid-resistant TB among Fulton County’s homeless population began in 2008.[21] As at December 2017, a total of 110 TB cases in the county have been linked to this outbreak.[22] In the report by Powell et al, the epidemiological curve of this outbreak was described in three phases–initial phase when only one facility was involved (2008–2009), quiescence phase when cases waned (2010–2013) and the recrudescence phase when multiple facilities became involved (2014 and thereafter).[21,22] A full description of the epidemiology of the associated cases and the outcomes of the direct contact tracing are well documented in prior studies.[21,22]
Following the resurgence of the outbreak in 2014, several county-wide outbreak control measures were instituted.[21,22] One of these measures was the implementation of mandatory TB screening in all Atlanta homeless housing facilities in May 2015.[21,22] Prior to this, outbreak-related TB screening services were only provided onsite at the Fulton County health department or at sporadic intervals on mobile units sent out to the affected shelters. Acceptance of TB screening was however left to the discretion of the shelter user. With the new TB screening policy, TB screenings were made available several days every week at many homeless shelter locations, whether an outbreak-related case had been identified at the shelter or not. The screenings were also made mandatory for all persons seeking use of these facilities, irrespective of their direct contact status with a previously diagnosed outbreak TB case. Each shelter user was expected to get screened for TB within 7 days of initial shelter entry and every 6 months thereafter to guarantee continued access to any of the shelters.[22] Both onsite (TB clinic) and offsite (shelter locations) TB screenings were performed using either tuberculin skin test (TST) or an interferon gamma release assay method (QuantiFERON Gold-in-tube tests (QFT)).[21,22] Homeless persons who tested positive for TB infection either through TST (>10mm induration) or QFT were referred to the health department for clinical evaluation. Homeless persons with positive QFT or TST who presented at the TB clinic for evaluation were clinically assessed to exclude active TB disease and those diagnosed with untreated LTBI were offered preventive treatment with either four months of rifampin (4R) (if patient had over 2 weeks history of consistent homeless shelter stay) or three months of Rifapentine-Isoniazid (3HP) ((if they had lower than 2 weeks of consistent homeless shelter stay). [22]
Another outbreak control measure implemented after the resurgence was in the area of LTBI treatment among homeless persons.[22] Prior to this, all LTBI treatment were self-administered (SAT). With the support of the Georgia State Department of Public Health, the Atlanta TB taskforce instituted the use of directly observed treatment (DOT) for LTBI treatment among homeless persons undergoing treatment with the health department. DOT was provided 5 days a week for homeless persons on rifampin and once a week for those on Rifapentine-isoniazid treated persons.
This analysis was limited to homeless persons who were treated for LTBI at the county health department using rifampin between January 2014 and December 2016.
LTBI treatment groups
Homeless persons treated by SAT were provided a pill box containing 30 doses of rifampin each month to self-administer daily. Self-reports of treatment adherence were recorded for SAT subjects at each monthly follow-up visit. Those treated by DOT met with trained disease intervention specialists each business day of the week at pre-determined locations (current shelter of residence, the park or the TB clinic) to receive their daily dose of rifampin. Assigned doses were ingested under direct observation and daily logs of successfully administered doses were maintained by the disease intervention specialists. Data on patient characteristics at treatment initiation including demographics (gender, age and race), clinical history (HIV status, self-reported social characteristics (any current alcohol use, non-alcoholic substance use (injectable and non-injectable) and mental/behavioral health-diagnosis in the past) were measured.
Treatment outcome
The outcome of interest was treatment completion. For patients in the SAT group, treatment completion was defined as picking up all four months’ supply of rifampin within five months of treatment initiation with self-reports of complete ingestion of all doses provided at the clearance visit. For the DOT-treated group, treatment completion was defined as taking all assigned doses of rifampin (5 doses per week for 16 weeks + 6 extra doses = 86 doses) within five months of treatment initiation.
Data analysis
Distributions of patient characteristics were calculated using frequencies and percentages. All patients treated with rifampin prior to May 2015 were treated by SAT while DOT was almost exclusively used after May 2015. Due to the very large proportion of study subjects who identified as Black or African American, all persons who were not African American (i.e. non-Hispanic whites, Hispanic/Latinos, Asians, Arabs, Native Americans, Pacific Islanders and others) were placed in a larger “Not Black” group to avoid having data counts less than 5 in some categories of race. To reduce bias due to non-random assignment to treatment group, propensity scores were used to balance the distribution of client characteristics in the treatment groups at baseline. Propensity score analysis is a statistical technique used to reduce bias resulting from non-random nature of treatment assignment seen in observational studies.[23–27] The estimated propensity scores (PS) represent the probability of treatment assignment conditional on the measured baseline covariates.[24–26] We estimated the propensity of each patient receiving the treatment type used to treat LTBI by modeling the treatment type received on all the measured baseline covariates in a logistic regression model.[25–27]. We examined the degree of overlap in the distributions of estimated propensity scores in the treatment groups (Common support) using histograms.[24,26,28]
Four methods of applying the scores in assessment of treatment effect (stratification, matching, inverse probability of treatment weighting using the propensity score, and covariate adjustment) have been described.[24–27,29,30] Among the four, weighting using inverse probability of treatment and propensity score matching have been shown to eliminate a greater degree of systematic differences between treatment arms.[29] While propensity score matching was shown to have marginally superior performance compared to weighting with inverse propensity of treatment, it requires the formation of matched sets of subjects with similar scores drawn from both treatment arms and may entail exclusion of patients when a comparable match is not available.[25,27,31,32] Given the small number of subjects in this study cohort (n = 274), we selected to use weighting with inverse probability of treatment to benefit from its superior performance in reducing confounding while retaining all observations in the analysis. The inverse of the estimated propensity scores was calculated and stabilized (to reduce variability in each group and reduce the influence of outlier weights) by multiplying the inverse of probability of treatment weights (IPTW) with the mean weights of the corresponding treatment group.[24–26,28,30] We assessed the degree to which confounding by the measure covariates have been balanced by comparing the standardized differences in the unweighted and weighted samples. Standardized difference <0.15 was used as the bench-mark to indicate balance of observed baseline characteristics after application of the weighted propensity scores.[24,27]
Independent associations between treatment completion and observed patient characteristics were assessed using unadjusted logistic regression models. The effect of modality used (DOT vs. SAT) on treatment completion was estimated using a weighted parsimonious logistic regression model (containing the main exposure alone) and also with a doubly robust logistic regression model (weighted logistic regression model that was fully adjusted for all measured covariates)[24,27,32–34]. For all statistical tests, p values <0.05 were considered to indicate significant associations. All statistical analyses were performed in SAS v 9.4 (SAS Institute, Cary, NC).
Results
TB screening and evaluation
A total of 13, 082 TB screening tests [TST– 9,462 (78%), QFT– 3,620 (28%)] were performed among 9,798 homeless persons between January 2014 and December 2016 (Fig 1). Seventy-eight percent (n = 7,688) of these persons were screened just once within the study period, 1,366 (14%) were screened twice and 744 (8%) were screened three or more times within the period (range 3–13 times). Nine percent of all individuals screened (926/9798) had a positive TB screening test results (by either TST or QFT) and were referred to the health department for further evaluation. Among those referred, 777 (84%) presented at the TB clinic for clinical evaluation and 149 persons did not follow up for evaluation. Eight persons among those who attended the referral appointment did not complete the evaluation process. Two of these had sputum samples collected for acid-fast bacilli smear and culture but the patients did not return (and could not be located) after the initial visit, one person signed a refusal for further evaluation and five left the clinic before a clinician could attend them.
Among the 769 persons who completed the evaluation process, 350 (46%) were diagnosed with untreated LTBI and were offered LTBI treatment. A total of 321 (92%) persons accepted to initiate treatment and 29 persons did not. Eighteen (6%) of those who accepted treatment were started on 3HP while 303 (94%) were placed on 4R regimen. Twenty-nine persons among the 4R initiators notified the TB clinic of their relocation from the area during treatment and had their treatment successfully transferred to other jurisdictions. These were excluded from the study, leaving 274 homeless persons in the final study cohort. Sixty-six percent (n = 181) of the 274 were treated by DOT while 93 (34%) were treated by SAT.
Study cohort
Of the 274 homeless persons who initiated daily rifampin for LTBI treatment at the health department, 89% were men, 93% were African American and 80% were over 39 years old (age range: 22–77 years). About 19% reported current alcohol use, 8% reported a history of illicit drug use and 9% reported living with at least one diagnosed mental health problem ranging from anxiety disorder to schizophrenia (Table 1). Three percent of the cohort was HIV positive. There were statistically significant differences in the gender (p = 0.008), alcohol use (p = 0.001), drug use (p = 0.036) and mental health diagnoses (p = 0.006) distribution of the homeless persons treated by DOT compared to SAT.
Propensity score estimation and diagnostics
All measured patient characteristics were included in the propensity score estimation model and histograms comparing the distribution of the estimated propensity scores showed considerable overlap in the spread of the scores in both treatment groups (Fig 2). Details of the covariate balance assessment are shown in Table 2. All measured characteristics were evenly distributed between the treatment groups after weighting with all standardized differences below the bench-mark (0.15).
LTBI treatment completion and bivariate analysis
Overall, 65% of the all 274 rifampin initiators (unweighted sample) completed LTBI treatment and a slightly higher proportion of DOT-treated patients (65%) than SAT-treated patients (63%) completed treatment. More alcohol users treated by DOT (62%) completed LTBI treatment than those treated by SAT (43%). In the (IPTW) weighted sample, treatment completion among the DOT-treated group was 68% while completion among the SAT-treated group was 62% (Table 3).
Details of bivariate analysis (unweighted sample) are shown in Table 4. Unadjusted analysis of the unweighted sample showed a small, non-significant improvement in treatment completion among the DOT-treated group compared to the SAT-treated group [unadjusted odds ratio (95% CI), cOR: 1.08 (0.64, 1.82), p = 0.774]. Among all covariates explored in unadjusted analysis, age between 50 and 59 years was the only patient characteristic marginally associated with treatment completion [cOR: 1.08 (1.96 (1.00, 3.83), p = 0.05].
Estimation of treatment effect
Adjusted analysis (Table 5) with the parsimonious, IPTW-weighted model showed that DOT significantly improved the odds of LTBI treatment completion in the sample by 30% compared to SAT ([Odds ratio (95% CI), OR: 1.30 (1.01, 1.67), p = 0.045]. Addition of all measured covariates in the doubly robust model (fully weighted and adjusted model) showed similar effect of DOT on treatment completion [OR (95% CI): 1.40 (1.07, 1.82), p = 0.014]. In this fully adjusted and weighted model, male gender, black race, age (50–59, 60 years and older) and positive HIV status were also significantly associated with increased odds of treatment completion while alcohol use was associated with decreased odds of completing treatment [OR (95% CI): 1.51 (0.36, 0.74)].
Discussion
Detection and treatment of LTBI among homeless persons is of immense public health importance in achieving the goal of TB elimination in the United States[3,5]. The presence of unaddressed co-morbidities (mental health-related and non-mental health-related), substance addictions (alcoholic and non-alcoholic) and poor access to medical care make effective treatment of LTBI in these persons difficult[11]. These factors in combination with poor adherence to LTBI treatment make the homeless population a fertile ground for frequent TB outbreaks.
In this retrospective study, about one third of the homeless persons who initiated rifampin for LTBI treatment completed their prescribed treatment. This compares favorably to the findings from the published studies on LTBI treatment adherence among homeless persons[18,35–38]. Goswami et al in their study reported treatment completion rate of 53% among homeless persons who initiated LTBI treatment while Nyamathi et al reported at overall completion rate of 51% among 520 homeless adults who initiated treatment for LTBI. The study by Tulsky et al reported a higher treatment completion (86%) in their randomized controlled trial on LTBI treatment among homeless persons in California.[36] However, only 49% of their study subjects were reported to be current residents of an emergency shelter at the time of treatment initiation[36] In comparison, all the subjects included in this study were current residents of an emergency shelter at the time of LTBI treatment initiation indicating a higher concentration of the target population in this study than in the study by Tulsky et al.
Propensity score analysis was employed in the assessment of the effect of treatment modality used on treatment completion. This was done primarily because there were statistically significant and clinically meaningful differences between the distributions of patient characteristics in the two treatment groups at treatment initiation. Patients who were treated by SAT were more likely to be men, non-reporters of alcohol or illicit drug use and have no history of previously diagnosed mental health disorder at treatment initiation than those treated by SAT. Appropriately specified propensity scores simulate the conditions obtained in randomized controlled trials, the gold standard of epidemiologic studies, and is usually used to achieve exchangeability of exposure groups in observational studies.[24,27,30,32,39] Examination of the spread of the estimated propensity scores showed considerable overlap in the distribution of the scores in both treatment groups indication appropriate specification of the propensity score estimation model. We also saw considerable reduction in the standardized differences between the covariate distributions in the weighted sample further indicating that balance in baseline distribution of the measured patent characteristics was achieved and the underlying propensity of treatment used for each subject in the cohort was appropriately specified. [24,27,30,39]
One of the core benefits of using propensity score based models to estimate treatment effect is the ability to model outcome using only one covariate (the main exposure) in a parsimonious model as against including multiple covariates in the effect estimate model in the traditional multivariate models. However, other variations to this have been used in scientific literature and this includes adding relevant variables in the propensity score outcome model to yield a model that is both fully adjusted and propensity score weighted (doubly robust) outcome model. [24,27,32–34] Both approaches (parsimonious and doubly robust models) were used in this study to examine the effect of using DOT versus SAT in treating homeless persons for latent TB infection. Both outcome models showed that the use of DOT for LTBI treatment resulted in significant improvement in treatment completion among the homeless persons compared to SAT. SAT is the current standard of practice for LTBI treatment in the general population. However, several issues come into play when this modality is employed in LTBI treatment among homeless persons These issues include potential lack of appropriate medication storage environment, possibility of pill box loss or theft, possibility of exchanging the medications for money or other necessities and often simple forgetfulness in the pursuit of more urgent survival needs. We are not aware of any previous study focused solely on LTBI treatment adherence in homeless persons on daily rifampin treatment, however previous studies on DOT and LTBI treatment in the general population and other high-risk groups have all reported increased adherence on DOT.[15,16,40] Our study provides support to the use of this adherence-improving intervention for homeless persons on daily rifampin regimen for LTBI treatment.
Nearly one fifth of the study sample reported alcohol use at the time of treatment initiation and a higher proportion of the self-reported alcohol users treated by DOT in both the unweighted and weighted samples completed the prescribed LTBI treatment compared to the alcohol users treated by SAT. Estimation of effect in the doubly robust model showed that persons who reported alcohol use were less likely to complete treatment compared to those who did not report any alcohol use. Regular use of alcohol and/or alcohol dependence, conditions quite common among homeless persons, do not only constitute strong risk factors for LTBI progression to active TB disease [3], but are also deterrents to treatment adherence among homeless persons.[15] This benefit of improvement in treatment adherence among homeless alcohol users treated by DOT may justify its use in LTBI treatment to both reduce risk of progression to TB disease and improve the likelihood of treatment completion.
Limitations
Our study had some limitations. Data used for this analysis were abstracted from routine clinical care records and therefore, relied heavily on the accuracy and completeness of the clinical care notes. Social characteristics (alcohol use illicit drug use and mental health diagnoses) examined were self-reported findings noted in the clinical records of patient-clinician encounters and thus depended both on the thoroughness of the clinician in inquiring about and recording these responses and accuracy of the patient’s response. In addition, there may have been social desirability bias influencing the patient responses to inquiries about alcohol and drug use and the impact of this bias on the outcome was not examined. The DOT regimen used in this study was administered five days a week. There is no published data on the efficacy of rifampin given five days a week in latent tuberculosis treatment. This dosing regimen was used given the prevailing circumstances at the time.[22]
This study would have benefitted from moderation analyses to assess the consistency of treatment effect across various subgroups of homeless persons treated for LTBI. Previous studies that have combined propensity score analyses with moderation analyses show that while it is more ideal to use subgroup-specific propensity scores for subgroup analyses (rather than propensity scores estimated in the full cohort), correctly specified propensity scores estimated in the full cohort remain valid within subgroups as long as the scores accurately reflect the underlying propensity of treatment for each subject and the subgroups are of sufficient size.[27,31–34,41] Our study sample did not have sufficient sample power at the sub-group level to allow robust estimation of treatment effect in moderation analyses and we suggest future research on this as an improvement of the findings on this study.
Public health implications
As the march towards TB elimination in the United States progresses, stronger attention will need to be paid to the detection and treatment of latent TB among high-risk populations like persons who reside in emergency homeless shelters and transitional housing facilities. Given the low levels of treatment acceptance, adherence and completion seen in previous years in this population, it is imperative now to seek more effective means of improving LTBI treatment adherence and completion among homeless persons and decrease risk of progression to active disease and emergence of TB outbreaks. Our study provides the statistical evidence to support the use of DOT to treat LTBI in unstably-housed persons in urban cities within the United States. As a next step, we suggest that the potential for achieving similar improvement in LTBI treatment adherence as observed in our study with in-person DOT needs to be evaluated using video DOT due to its’ attractive benefit of reduced cost of implementation and manpower, particularly for rural areas where in-person DOT might be difficult to implement.
Acknowledgments
The authors wish to acknowledge all TB staff at Fulton County Board of Health, Georgia Department of Public Health, the CDC, MercyCare Atlanta and Atlanta TB taskforce for their spirited efforts in TB control during the TB outbreak among homeless persons in Atlanta, Georgia.
References
- 1.
WHO | The top 10 causes of death. WHO. 2017; Available at: http://www.who.int/mediacentre/factsheets/fs310/en/.
- 2.
Organization WH. Global tuberculosis report 2016. Geneva: World Health Organization, 2016. Available at: files/14/9789241565394-eng.pdf.
- 3. Schmit KM. Tuberculosis—United States, 2016. MMWR. Morb. Mortal. Wkly. Rep. 2017; 66. Available at: https://www.facebook.com/CDCMMWR.
- 4.
Controlling Tuberculosis in the United States: Recommendations from the American Thoracic Society, CDC, and the Infectious Diseases Society of America.
- 5. Hill AN, Becerra JE, Castro KG. Modelling tuberculosis trends in the USA. Epidemiol. Infect. 2012; 140:1862–1872. Available at: https://www.cambridge.org/core/article/modelling-tuberculosis-trends-in-the-usa/A528E00C4B928A75401EA8D32F88BD43. pmid:22233605
- 6. Narasimhan P, Wood J, Macintyre CR, Mathai D. Risk factors for tuberculosis. Pulm Med 2013; 2013.
- 7.
WHO | Tuberculosis control in vulnerable groups. WHO. 2017; Available at: http://www.who.int/bulletin/volumes/86/9/06-038737/en/.
- 8. Shea KM, Kammerer JS, Winston CA, Navin TR, Horsburgh Jr., Robert C. Estimated Rate of Reactivation of Latent Tuberculosis Infection in the United States, Overall and by Population Subgroup. Am. J. Epidemiol. 2014; 179:216–225. Available at: pmid:24142915
- 9. CDC. CDC | TB | Treatment | Treatment Regimens for Latent TB Infection (LTBI). 2017; Available at: https://www.cdc.gov/tb/topic/treatment/ltbi.htm.
- 10. Sterling TR, Villarino ME, Borisov AS, Shang N, Gordin F, Bliven-Sizemore E, et al. Three Months of Rifapentine and Isoniazid for Latent Tuberculosis Infection. N. Engl. J. Med. 2011; 365:2155–2166. Available at: pmid:22150035
- 11. Yamin A, Bornstein E, Hensel R, Mohamed O, Kempker RR. Predictors of Latent Tuberculosis Infection Treatment After Introduction of a New Regimen: A Retrospective Cohort Study at an Inner City Clinic. Open Forum Infect. Dis. 2016; 3:ofw082. Available at: pmid:27757409
- 12. McClintock AH, Eastment M, McKinney CM, Pitney CL, Narita M, Park DR, et al. Treatment completion for latent tuberculosis infection: a retrospective cohort study comparing 9 months of isoniazid, 4 months of rifampin and 3 months of isoniazid and rifapentine. BMC Infect. Dis. 2017; 17:146. Available at: pmid:28196479
- 13. Eastment MC, McClintock AH, McKinney CM, Narita M, Molnar A. Factors That Influence Treatment Completion for Latent Tuberculosis Infection. J. Am. Board Fam. Med. 2017; 30:520–527. Available at: http://www.ncbi.nlm.nih.gov/pubmed/28720633. Accessed 14 December 2017. pmid:28720633
- 14. Lobato MN, Reves RR, Jasmer RM, Grabau JC, Bock NN, Shang N, et al. Adverse events and treatment completion for latent tuberculosis in jail inmates and homeless persons. Chest 2005; 127:1296–1303. Available at: http://www.sciencedirect.com/science/article/pii/S0012369215344809. pmid:15821208
- 15. Stuurman AL, Vonk Noordegraaf-Schouten M, van Kessel F, Oordt-Speets AM, Sandgren A, van der Werf MJ. Interventions for improving adherence to treatment for latent tuberculosis infection: a systematic review. BMC Infect. Dis. 2016; 16:257. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4897858/. pmid:27268103
- 16. Sandgren A, Vonk Noordegraaf-Schouten M, van Kessel F, Stuurman A, Oordt-Speets A, van der Werf MJ. Initiation and completion rates for latent tuberculosis infection treatment: a systematic review. BMC Infect. Dis. 2016; 16:204. Available at: pmid:27184748
- 17. Vermeire E, Hearnshaw H, Royen V, P, Denekens J. Patient adherence to treatment: three decades of research. A comprehensive review. J. Clin. Pharm. Ther. 2001; 26:331–342. Available at: http://onlinelibrary.wiley.com/doi/10.1046/j.1365-2710.2001.00363.x/full. pmid:11679023
- 18. Nyamathi A, Nahid P, Berg J, Burrage J, Christiani A, Aqtash S, et al. Efficacy of Nurse Case-Managed Intervention for Latent Tuberculosis Among Homeless Subsamples. Nurs. Res. 2008; 57:33–39. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3100804/. pmid:18091290
- 19. Malotte CK, Hollingshead JR, Larro M. Incentives vs outreach workers for latent tuberculosis treatment in drug users. Am. J. Prev. Med. 2001; 20:103–107. Available at: http://www.ajpmonline.org/article/S0749-3797(00)00283-X/abstract. pmid:11165450
- 20. Hirsch-Moverman Y, Daftary A, Franks J, Colson PW. Adherence to treatment for latent tuberculosis infection: systematic review of studies in the US and Canada. Int. J. Tuberc. Lung Dis. 2008; 12:1235–1254. Available at: http://www.ingentaconnect.com/content/iuatld/ijtld/2008/00000012/00000011/art00005. pmid:18926033
- 21. Powell KM, VanderEnde DS, Holland DP, Haddad MB, Yarn B, Yamin AS, et al. Outbreak of Drug-Resistant Mycobacterium tuberculosis Among Homeless People in Atlanta, Georgia, 2008–2015. Public Health Rep. 2017; 132:231–240. Available at: pmid:28257261
- 22. Holland DP, Alexander S, Onwubiko U, Goswami ND, Yamin A, Mohamed O, et al. Response to Isoniazid-Resistant Tuberculosis in Homeless Shelters, Georgia, USA, 2015–2017. Emerg. Infect. Dis. 2019; 25:593–595. Available at: http://wwwnc.cdc.gov/eid/article/25/3/18-1678_article.htm. Accessed 1 March 2019. pmid:30789335
- 23. Polkinghorne KR. Vascular Access and All-Cause Mortality: A Propensity Score Analysis. J. Am. Soc. Nephrol. 2004; 15:477–486. Available at: pmid:14747396
- 24. Vaughan AS, Kelley CF, Luisi N, del Rio C, Sullivan PS, Rosenberg ES. An application of propensity score weighting to quantify the causal effect of rectal sexually transmitted infections on incident HIV among men who have sex with men. BMC Med. Res. Methodol. 2015; 15:25. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4369368/. pmid:25888416
- 25. Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behav. Res. 2011; 46:399–424. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3144483/. pmid:21818162
- 26. Lanehart RE, de Gil PR, Kim ES, Bellara AP, Kromrey JD, Lee RS. Propensity score analysis and assessment of propensity score approaches using SAS procedures. 2012. Available at: https://www.researchgate.net/profile/Rheta_Lanehart/publication/275831465_PROPENSITY_SCORE_ANALYSIS_AND_ASSESSMENT_OF_PROPENSITY_SCORE_APPROACHES_USING_SAS_R_PROCEDURES/links/5547c6620cf2b0cf7ace9434/PROPENSITY-SCORE-ANALYSIS-AND-ASSESSMENT-OF-PROPENSITY-SCORE-APPROACHES-USING-SAS-R-PROCEDURES.pdf.
- 27. Rosenbaum PR, Rubin DB. The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika 1983; 70:41–55. Available at: http://www.jstor.org/stable/2335942.
- 28. Garrido MM, Kelley AS, Paris J, Roza K, Meier DE, Morrison RS, et al. Methods for Constructing and Assessing Propensity Scores. Health Serv. Res. 2014; 49:1701–1720. Available at: pmid:24779867
- 29. Austin PC. The Relative Ability of Different Propensity Score Methods to Balance Measured Covariates Between Treated and Untreated Subjects in Observational Studies. Med. Decis. Mak. 2009; 29:661–677. Available at: http://www.ncbi.nlm.nih.gov/pubmed/19684288. Accessed 1 March 2019.
- 30. Harder VS, Stuart EA, Anthony JC. Propensity score techniques and the assessment of measured covariate balance to test causal associations in psychological research. Psychol. Methods 2010; 15:234–249. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2936698/. pmid:20822250
- 31. Liem YS, Wong JB, Hunink MM, de Charro FT, Winkelmayer WC. Propensity scores in the presence of effect modification: A case study using the comparison of mortality on hemodialysis versus peritoneal dialysis. Emerg. Themes Epidemiol. 2010; 7:1. Available at: Accessed 1 March 2019. pmid:20459823
- 32. Elze MC, Gregson J, Baber U, Williamson E, Sartori S, Mehran R, et al. Comparison of Propensity Score Methods and Covariate Adjustment: Evaluation in 4 Cardiovascular Studies. J. Am. Coll. Cardiol. 2017; 69:345–357. Available at: https://www.sciencedirect.com/science/article/pii/S073510971637036X. Accessed 5 March 2019. pmid:28104076
- 33. Kang JDY, Schafer JL. Demystifying Double Robustness: A Comparison of Alternative Strategies for Estimating a Population Mean from Incomplete Data. Stat. Sci. 2007; 22:523–539. Available at: http://projecteuclid.org/euclid.ss/1207580167. Accessed 6 March 2019.
- 34. Green KM, Stuart EA. Examining moderation analyses in propensity score methods: application to depression and substance use. J. Consult. Clin. Psychol. 2014; 82:773–83. Available at: http://www.ncbi.nlm.nih.gov/pubmed/24731233. Accessed 5 March 2019. pmid:24731233
- 35. Goswami ND, Gadkowski LB, Piedrahita C, Bissette D, Ahearn MA, Blain ML, et al. Predictors of latent tuberculosis treatment initiation and completion at a U.S. public health clinic: a prospective cohort study. BMC Public Health 2012; 12:468. Available at: http://www.ncbi.nlm.nih.gov/pubmed/22720842. Accessed 29 October 2018. pmid:22720842
- 36. Tulsky JP, Hahn JA, Long HL, Chambers DB, Robertson MJ, Chesney MA, et al. Can the poor adhere? Incentives for adherence to TB prevention in homeless adults. Available at: http://www.ingentaconnect.com/content/iuatld/ijtld/2004/00000008/00000001/art00012;jsessionid=kj9ae1p5p7wc.x-ic-live-03#. Accessed 29 October 2018.
- 37. Lashley M. A Targeted Testing Program for Tuberculosis Control and Prevention Among Baltimore City’s Homeless Population. Public Health Nurs. 2007; 24:34–39. Available at: Accessed 29 October 2018. pmid:17214651
- 38. Stout JE, Engemann JJ, Cheng AC, Fortenberry ER, Hamilton CD. Safety of 2 Months of Rifampin and Pyrazinamide for Treatment of Latent Tuberculosis. Am. J. Respir. Crit. Care Med. 2003; 167:824–827. Available at: http://www.ncbi.nlm.nih.gov/pubmed/12446275. Accessed 29 October 2018. pmid:12446275
- 39. Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat. Med. 2015; 34:3661–3679. Available at: pmid:26238958
- 40. Perlman DC, Gourevitch MN, Trinh C, Salomon N, Horn L, Des Jarlais DC. Cost-Effectiveness of Tuberculosis Screening and Observed Preventive Therapy for Active Drug Injectors at a Syringe-Exchange Program. 2001. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3455907/pdf/11524_2006_Article_51.pdf. Accessed 29 October 2018.
- 41. Rassen JA, Glynn RJ, Rothman KJ, Setoguchi S, Schneeweiss S. Applying propensity scores estimated in a full cohort to adjust for confounding in subgroup analyses. Pharmacoepidemiol. Drug Saf. 2012; 21:697–709. Available at: http://www.ncbi.nlm.nih.gov/pubmed/22162077. Accessed 5 March 2019. pmid:22162077