Interventions for improving adherence to treatment for latent tuberculosis infection: a systematic review

Background Latent tuberculosis infection (LTBI) control relies on high initiation and completion rates of preventive treatment to preclude progression to tuberculosis disease. Specific interventions may improve initiation and completion rates. The objective was to systematically review data on determinants of initiation, adherence and completion of LTBI treatment, and on interventions to improve initiation and completion. Methods A systematic review of the literature (PubMed, Embase) published up to February 2014 was performed. Relevant prospective intervention studies were assessed using GRADE. Results Sixty-two articles reporting on determinants of treatment initiation and completion were included and 23 articles on interventions. Determinants of LTBI treatment completion include shorter treatment regimen and directly observed treatment (DOT, positive association), adverse events and alcohol use (negative association), and specific populations with LTBI (both positive and negative associations). A positive effect on completion was noted in intervention studies that used short regimens and social interventions; mixed results were found for intervention studies that used DOT or incentives. Conclusion LTBI treatment completion can be improved by using shorter regimens and social interventions. Specific needs of the different populations with LTBI should be addressed taking into consideration the setting and condition in which the LTBI treatment programme is implemented. Electronic supplementary material The online version of this article (doi:10.1186/s12879-016-1549-4) contains supplementary material, which is available to authorized users.


Background
Exposure to Mycobacterium tuberculosis may result in latent tuberculosis infection (LTBI). LTBI can in turn progress to tuberculosis (TB) disease, especially if the immune system is compromised [1,2]. One-third of the world population is estimated to be latently infected with M. tuberculosis [3]; therefore, LTBI control is an important step towards TB elimination, in addition to TB case detection and treatment [4,5]. LTBI control consists of identifying individuals with LTBI and offering them preventive treatment. The fact that initiation and completion rates of preventive treatment are often low and differ between treatment regimens and populations with LTBI may hamper the control of LTBI.
Numerous factors can influence patients' medication uptake, such as forgetfulness, side effects, stigma, or lack of information on treatment requirements, thereby affecting initiation, adherence and completion rates of treatment. These factors affecting patients' medication uptake have to be considered when designing interventions to modify complex human behaviour associated with treatment adherence [6][7][8]. Specific examples of interventions that have been used to improve initiation, adherence, or completion of LTBI treatment include switching from regimens with longer treatment duration to regimens with shorter treatment duration [9], incentives [10], and education or counselling [7].
In order to provide European Union and European Economic Area (EU/EEA) Members States and candidate countries with guidance on programmatic LTBI control a systematic review on initiation and completion of preventive treatment was performed in a bilateral cooperation with the World Health Organization (WHO) [11]. The review served as input for the WHO guidelines on management of LTBI launched in 2015 [11], and will be used for the EU/EEA tailored guidance. The review questions were (1) What are the initiation and completion rates of different LTBI treatment regimens?; (2) What are the determinants of initiation, adherence, and completion of recommended LTBI treatment regimens in the general population and in specific populations with LTBI?; and (3) What are interventions with demonstrated efficacy to improve LTBI treatment initiation, adherence and completion in different populations?. The current article presents the results on the latter two review questions.

Methods
A systematic literature review was performed to provide answers to the review questions described above. This review was done according to a review protocol and following the Cochrane guidelines. The details of eligibility criteria, information sources, search strategy, study selection, and data extraction are provided in the Additional files (see Additional file 1).

Data extraction
To answer review question 2, data on determinants of initiation, adherence, and completion were extracted for individuals with LTBI from various populations with LTBI (e.g. p-values, odds ratios (ORs), risk ratios). If univariate and multivariate analyses of the same data were presented, only the data from the multivariate analysis were extracted. Results from intention-to-treat analyses were preferred; if both intention-to-treat and per-protocol results were reported in one study, only intention-to-treat results were included. Data on non-significant factors were not consistently quantified in the studies and were therefore not listed in this review.
For review question 3, data for five groups of interventions were extracted: 1. interventions with short treatment regimens; 2. interventions consisting of directly observed therapy (DOT); 3. interventions in which incentives were offered with the treatment (e.g. cash, transportation vouchers); 4. social interventions (e.g. education, adherence coaching, peer counselling, cultural interventions); and 5. other interventions (e.g. use of interferon gamma release assay (IGRAs) rather than tuberculin skin tests (TSTs)). ORs that were adjusted for factors that related to the intervention were not used.

GRADE
As review question 2 does not deal with the effects of health interventions, risk of bias was only assessed for aspects of the individual studies and not across the evidence base. For review question 3, the quality of the total body of evidence for each outcome (initiation, adherence, and completion rates) was critically appraised using the GRADE approach (Grading of Recommendations Assessment, Development and Evaluation; http://www.gradeworkinggroup.org). Only prospective studies (i.e. randomized controlled trials (RCTs) and prospective observational studies) were appraised using GRADE. Outcomes were downgraded for imprecision when the total number of events was less than 125 for dichotomous outcomes, and if the total sample size was less than 230 for continuous outcomes (based on estimated control group event rate of 0.60 and a relative risk difference of 30 %, given α = 0.05 and β = 0.20) [12]. GRADE tables were created using standard GRADE formats and procedures (with GRADEpro [13]). Metaanalysis was performed in accordance with GRADE methodology: summary odds ratios (sORs) and 95 % confidence intervals were calculated when the outcomes were considered relatively homogeneous regarding the intervention and the population. This was done using a random effects model with the "MAInput Table" and "MAPooledEffect" functions from the MetaXL 2.1 add-in in Excel. No quality index was used. The sORs and a measure for heterogeneity (I 2 ) are shown in a forest plot [14].

Results of the review process
Results are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. A flowchart showing the number of articles identified for all review questions during the selection process is presented in in the Additional files (see Additional file 2).
Overall, 62 articles were found for review question 2, including 27 prospective studies and 35 retrospective studies (see Additional file 3). Twelve articles provided information on determinants of initiation, eight on determinants of adherence, and 51 on determinants of completion. Twenty-three articles were found for review question 3, including twenty prospective and three retrospective studies (see Additional file 4). Among the prospective studies, seven articles described interventions with short treatment regimens, four with DOT, four with incentives, eight with social interventions, and one described another type of intervention. Four prospective studies provided data for more than one outcome [15][16][17][18][19][20][21].

Determinants of LTBI treatment initiation, adherence and completion
Most determinants of LTBI treatment initiation, adherence, and completion are from studies in the general population, i.e. primarily unselected patients with LTBI at clinics ( Table 1). The most frequently reported determinant associated with LTBI treatment uptake in the general population was age, though the direction of the effect was inconsistent. Most determinants related to LTBI treatment completion.
With regards to LTBI treatment initiation, two studies found healthcare workers (vs. non healthcare workers) to be less likely to initiate treatment [22,23] (no study found a positive association); three studies found case contacts (vs. no case contacts) [22][23][24], and two studies found immigrants or refugees (vs. born in country of study) [25,26] to be more likely to initiate LTBI treatment (no study found an inverse association). With regards to LTBI treatment completion, five studies found a positive association between completion and immigrant or refugee status (vs. born in country of study) [27][28][29][30][31], whereas two studies found an inverse association [32,33]. Two studies each found that currently homeless individuals [30,32] (vs. not currently homeless) and people who injected drugs (PWID) [23,34] (vs. people who do not inject drugs) were less likely to complete treatment (no study found a positive association). Throughout the populations, unfavourable social risk factors were associated with worse completion [23,25,27,[35][36][37][38] (no study found a positive association). Determinants of treatment initiation, adherence or completion within specific source populations with LTBI are presented in the Additional files (see Additional file 3), and showed large analogy with the determinants identified for the general population. Short (vs. long) treatment regimens and treatments with DOT (vs. self-administered therapy (SAT)) were found to be completed more often in the general population with LTBI in six [31,[39][40][41][42][43] and three [31,44,45] studies, respectively, (no study found an inverse association). Adverse events were inversely associated with completion in seven studies [30,33,41,[46][47][48][49] (no study found a positive association). Similar results were found for determinants of completion in individuals from the different populations (see Additional file 3). Females [30,44] (vs. males) were more likely to complete treatment (no study found men were more likely to complete treatment). Additionally, alcohol use (vs. no alcohol use) was inversely associated with completion in four studies [29,30,32,49] (no study found a positive association).

Interventions to improve LTBI treatment initiation, adherence and completion
Twenty prospective studies on interventions to improve LTBI treatment initiation, adherence, and/or completion provided evidence for the following questions (Tables 2, 3, 4, 5, 6): Does short LTBI treatment result in higher initiation, adherence, or completion rates than long LTBI treatment in individuals eligible for LTBI treatment ( Table 2)?
Does DOT result in higher initiation, adherence, or completion rates than SAT in individuals eligible for LTBI treatment (Table 3)?
In undocumented migrants, significantly lower completion rates were found among those receiving twice weekly clinic-based DOT compared to daily SAT (OR = 0.1; 95 % CI 0.0-0.3; low quality of evidence) [54] or twice weekly SAT (OR = 0.2; 95 % CI 0.1-0.6). In one other study, in PWID, no effect of DOT administered by an outreach nurse on completion rates of LTBI treatment was found (OR = 1.1; 95 % CI 0.5-2.1; moderate quality of evidence). However, when looking at the proportion of people who took all doses, the DOT group performed significantly better (OR = 31.5; 95 % CI 14.1-70.6) [17].
Two more studies compared DOT to SAT, but were confounded: among case contacts with LTBI a shorter treatment regimen was given in the DOT than in the SAT group [20], and among PWID with LTBI the DOT group received methadone treatment whereas the SAT group did not [16]. Higher completion rates were found in the DOT group among both case contacts (OR = 2.1; 95 % CI 1.9-2.3; low quality of evidence) and PWID (OR = 14.5; 95 % CI 5.1-42; very low quality of evidence).
Does treatment supported by (monetary) incentives result in higher initiation, adherence, or completion rates than treatment not supported by incentives in individuals eligible for LTBI treatment (Table 4)?
Two studies in PWID with LTBI found higher completion rates for LTBI treatment among those who received either a monetary incentive (adjusted OR [aOR] = 32.0; Missed appointment call or letter (vs. no missed appointment call) No medical insurance (vs. medical insurance) ---1 [47] Clinic attendance before treatment (vs. clinic non-attendance before treatment) -

[79]
--95 % CI 7.1-145; moderate quality of evidence) [55] or methadone treatment (OR = 14.5; 95 % CI 5.0-42; very low quality of evidence) [16] compared to those who received no incentive. The results from the methadone treatment study were confounded, however, by the use of DOT in the methadone treatment group and SAT in the control group. The provision of food or transportation vouchers to released inmates with LTBI if they attended a TB clinic upon release (OR = 1.1; 95 % CI 0.5-2.4; moderate quality of evidence) [15] did not lead to better completion rates. In another study, no difference was found between the provision of cash-incentives versus non-cashincentives to homeless individuals with LTBI (OR = 1.7; 95 % CI 0.7-4.3; low quality of evidence) [56].
Do social interventions result in higher initiation, adherence, or completion rates than standard care in individuals eligible for LTBI treatment (Table 5)?
Adherence coaching among the general population with LTBI at clinics (low quality of evidence) and a cultural intervention among immigrants (very low quality of evidence) resulted in better adherence [19,57]. Social interventions were found to improve completion rates of LTBI treatment compared to the standard care group in all but one study, which provided peer-support among PWID with LTBI (OR = 1.0; 95 % CI 0.7-1.5; high quality of evidence) and found no effect on completion [17]. Counselling and contingency contracting, adherence coaching and self-esteem counselling, and peer-based interventions in the general population showed better completion rates (sOR = 1.4; 95 % CI 1.1-19; low heterogeneity, high quality of evidence) [19,27,58] (Fig. 3). Education among inmates (OR = 2.2; 95 % CI 1.0-4.7; moderate quality of evidence) [15], nurse case management among homeless individuals (aOR = 3.0; 95 % CI 2.2-4.2; high quality of evidence) [35], and case management with attention for the cultural background of each individual among immigrants (aOR = 7.8; 95 % CI 5.7-10.7; low quality of evidence) [18] improved completion. The latter study also found that this intervention led to higher initiation rates (OR = 2.7; 95 % CI 1.9-3.8; low quality of evidence).
Do interventions other than short treatment, directly observed therapy, incentives or social interventions result in higher initiation, adherence, or completion rates than standard care in individuals eligible for LTBI treatment ( Table 6)?
One study showed that the use of IGRAs rather than TSTs for diagnosis of LTBI improved the initiation rate of LTBI treatment in healthcare workers with LTBI (OR = 8.8; 95 % CI 3.1-23; very low quality of evidence) [59].
Three retrospective studies performed in the general population with LTBI showed significantly higher completion rates in the groups that received DOT, behaviour modification techniques in combination with incentives, or home to clinic follow-up, respectively [28,44,60] (see Additional file 4).

Discussion
To the best of our knowledge, this is the first review to systematically and comprehensively summarise data on any determinant of, and intervention to improve, LTBI treatment initiation, adherence and completion in all types of populations.

Determinants
The most frequently found determinants of treatment completion in our review were patient-related (i.e. type of population with LTBI, demographic factors, drug/alcohol abuse), therapy-related (e.g. short therapy regimens, DOT, occurrence of adverse events), and socioeconomic (e.g. unemployment, lack of social support). Unfavourable socio-economic factors were consistently associated with poor completion of LTBI therapy. These results should be interpreted with care, since     [20]: unclear allocation concealment; no blinding; use of unvalidated patient-reported outcomes in SAT arm (pill count and self-report); dissimilarities between treatment arms (with respect to North American Indians, subjects enrolled in a cluster, homelessness); exposure bias (short treatment in DOT arm) h Matteelli et al. 2000 [54]: unclear allocation concealment; no blinding; very large loss to follow-up; unclear treatment adherence assessment in SAT arm; unequal numbers in treatment arms; early termination (due to low completion rates in DOT arm). Early termination partially accounts for the low numbers in this study, and as we already downgraded for this (serious imprecision), we decided not to downgrade for it again in the risk of bias    [56]: partly blinded; dissimilarities between treatment arms (primary housing in last year shelter/street; not found to be an independent predictor of completion in this study)this study presents data for incentive vs. another incentive (rather than vs. no incentive) j Adjusted OR, not reported which factors this OR was adjusted for k $25 equivalent in food or transportation vouchers l Total number of events <125 m Patients with normal chest X-rays prescribed H, while those with evidence of old TB on chest X-ray were prescribed HR. Participants randomly assigned to the cash or non-cash incentive. Non-cash incentives consisted of a choice of $5 equivalent in fast-food or grocery store coupons, phone cards or bus tokens   Goldberg et al. 2004 [18]: use of unvalidated patient-reported outcomes (self-report); proportion of children aged 5-14 years was higher during one period than the other (19 % vs. 13 %, p = 0.003) f No adherence rates were provided as outcome; instead, the cumulative mean number of pills taken per group was presented  [35]: unclear allocation concealment; unclear sequence generation; partly blinded; dissimilarities between treatment arms (daily alcohol or drug use [significantly associated with non-completion in this study]; male, recruitment site [both not significantly associated with completion in this study], lifetime intravenous drug use, recent self-help program) p Adjusted OR, adjusted for: age, sex, high-school graduate, never married, medical insurance, recruited from homeless shelter, years homeless, treatment completion important, intended to adhere, daily alcohol/drug use, recent self-help program, emotional well-being, social support, recent hospitalization, recent victimization q Chaisson et al. 2001 [17]: unclear allocation concealment; no blinding; use of unvalidated patient-reported outcomes (self-report; urine tests and MEMS in a subset of patients in this study show that self-report is subject to serious under-reporting)  different measures of associations were used in the studies, the reference groups varied between studies, and data on non-significant factors were not always quantified in the studies and were therefore not listed in this review. However, the same socio-economic factors were also predictors of non-adherence to highly active antiretroviral therapy in human immunodeficiency virus (HIV)-patients or to cardiovascular medication [61][62][63][64][65]. Also, adverse events have been associated with worse adherence to treatment for all these conditions [61,62]. Similar factors were associated with adherence to treatment for active TB [61].

Interventions Initiation
Some evidence was found that the use of IGRAs rather than TSTs [59], or a social intervention using case management with attention to an individual's cultural background might positively influence the initiation rate of LTBI treatment [18].

Adherence
Our meta-analysis showed that case contacts had better adherence if they received short treatment compared to those on long treatment regimens [21,50]. Social interventions in the form of adherence coaching of adolescents with LTBI and cultural interventions among immigrants with LTBI also resulted in improved adherence [19,57].

Completion
Overall, completion rates of LTBI treatment were better among groups receiving shorter regimens than those with longer treatment regimens. The only outcome regarding the effect of shorter treatment on completion rates for which no effect was found could be explained by a relatively high rate of hepatotoxicity (11 %) found in the short treatment arm compared to the long treatment arm (3 %) [21]. This led to premature termination of the study. The applied short regimen of rifampicin plus pyrazinamide is currently not generally offered to persons with LTBI due to its association with hepatotoxicity [66]. Mixed results were found on the effect of DOT on completion rates of LTBI treatment. The significantly lower completion rates among those receiving clinic-based DOT in one study might be attributable to the difficulty undocumented migrants have in reporting regularly to health services to collect their drugs [54]. The study in which no effect was found among PWID did find that more people in the DOT group took all doses and, importantly, also  Fig. 2 Forest plot for completion of short vs. long LTBI treatment in the general population diagnosed with LTBI found that the number of self-reported doses taken was likely to be greatly over-estimated [17]. With regards to two other types of long-term treatment, highly active antiretroviral therapy in HIV-patients and TB treatment, no overall benefit of DOT compared to SAT was found on viral load among HIV-patients or cure among TB patients in reviews by Nachega et al. and Volmink et al. [67,68].
Of the studies reporting on the effect of incentives, two studies conducted in PWID with LTBI found a positive result (one of which was confounded), and the other study in released inmates found no effect. The success of incentives is likely to be both population and setting dependent. Lutge et al. reviewed the literature on material incentives and treatment for latent TB or active TB disease and concluded that the effect on long-term adherence and completion is not clear [69].
Social interventions to improve LTBI treatment uptake included case management with attention for an individual's cultural background, adherence coaching, counselling, contingency contracting, education, nurse case management and peer-based interventions. Most studies on this topic showed better completion rates in the intervention group than in the standard-care group, regardless of the type of social intervention. In a review of RCTs, Schroeder et al. found patient education to be largely unsuccessful in improving adherence to blood pressure-lowering medication, whereas some motivational strategies and complex interventions were successful [70]. In three out of the seven RCTs in a review of Schedlbauer et al. social interventions (i.e. patient information and education, intensified patient care, or a complex behavioral approach) improved adherence rates to lipid-lowering medication) [71].

Other reviews
Several other reviews present data on interventions to improve medication uptake among LTBI patients. These reviews had specific questions, for example the effect of lay healthcare workers on completion of LTBI treatment [72], interventions to improve the health of the homeless [73], education or counselling to improve completion of LTBI treatment [7], effects of rifampicin monotherapy or rifamycin-combination therapy versus isoniazid for preventing active TB and the role of completion rates [9]; incentives to reinforce medication adherence, including for LTBI treatment [10,69]. However, these reviews included only a small number of studies with the specific aim to investigate interventions to improve LTBI treatment initiation, adherence or completion. Some of these articles were also included in the current review; some could not be included in our review because we used different inclusion criteria.

Limitations
The definitions of completion varied between the included studies, as did the ways in which treatment adherence was assessed and in which completion rates were calculated. For example definitions used varied from "completed four months of rifampicin" to "picked up nine months of isoniazid within twelve months" and "took at least 80 % of the prescribed medication within twenty weeks". This heterogeneity complicates comparison of rates between studies and hampers meta-analysis and interpretation of the results. Since adherence and completion are similar concepts in the sense that full adherence to a treatment regimen leads to its completion, the limitations that are applicable to measures for completion are also applicable to adherence.
There is no standard definition for LTBI using tuberculin reactivity which is universally accepted.
There were no pre-set LTBI diagnostic criteria for inclusion of the studies in this review, the inclusion relied on reporting of the diagnostic criteria of the individual studies; if the study considered a case to be diagnosed then the study was included in the review and those cases were analysed.
Only determinants that showed a statistically significant association with initiation, adherence and completion were listed in this article, this should be taken into account when interpreting the results. The power of a study to  Fig. 3 Forest plot for completion of LTBI treatment using social interventions in the general population diagnosed with LTBI detect a significant effect was not taken into account in this review. Additionally, the determinants were merely described and no summary analyses were done because the non-significant determinants were not quantified and because of heterogeneity between the included studies. Comparison of studies is also complicated by the fact that different measures of association were used by the included studies; reference groups may differ; or the definition of the determinant itself might vary. These intricacies would be lost when grouping the determinants.
We calculated summary estimates for studies with similar populations and interventions. However, the studies were still quite heterogeneous. Calculation of summary estimates by combining studies without correcting for possible bias-causing factors between studies may cause bias in the results. Furthermore, the I 2 estimates need to be interpreted with caution because each meta-analysis only included three studies.
Finally, when conducting a systematic review, any limitations of included studies (e.g. lack of controlling for relevant covariates) inherently become limitations of data presented in the review.

Gaps and future research
The number of intervention studies in specific populations was scarce. In order to generate evidence on the effectiveness of context-specific interventions to improve the uptake of LTBI treatment, RCTs tailored to specific populations, with consideration of available resources and infrastructure of the health system, are necessary [11]. Although clinically relevant, no determinants of initiation, adherence or completion were found for patients with comorbidities.
Ultimately, the effect of LTBI treatment on the development of TB disease is important, and initiation, adherence and completion rates of LTBI treatment are intermediate determinants. In a subgroup analysis of a network meta-analysis of RCTs to determine the most efficacious regimen for preventing active TB disease [74], no evidence of a relationship between adherence (expressed as overall percentage of doses received) and efficacy was found. Still, more information on the association between treatment adherence and efficacy for prevention of active TB may be valuable.

Conclusions
Clinical benefit to individuals with LTBI and the success of the LTBI control programme in general are dependent on individuals taking the medication and completing the full course of treatment [11]. In the first part of this review, it was found that initiation and completion rates of LTBI treatment were frequently suboptimal and varied greatly within and across different populations. Taking determinants of initiation, adherence, and completion into account is an important first to step to plan interventions to improve these rates.
The available evidence on the effect of interventions on treatment initiation, adherence and completion presented in this review suggests that some interventions, notably the use of shorter treatment regimens and social interventions, have a positive effect on adherence and completion.
Overall, however, the evidence was inconclusive and recommendations on the best interventions to improve uptake of LTBI medication are hampered by the heterogeneity of the studies. The benefit of interventions to improve treatment completion, such as incentives and DOT, appears to be population and setting dependent. Specific needs of the different populations with LTBI should be addressed taking into consideration the local context, specific settings and conditions in which the LTBI treatment programme is implemented.

Ethics approval and consent to participate
Not applicable.