Systematic review of risk factors for nonadherence to TB treatment in immigrant populations

Abstract

Background

Foreign-born populations carry a significant TB burden in low-prevalence countries, composing over half of all cases in parts of Europe and North America. This study systematically reviewed evidence of risk factors for nonadherence to TB drug therapy in this group.

Methods

On 28 October 2013 MEDLINE, CINAHL, Embase, PsychINFO and ProQuest were systematically searched for studies examining adherence in foreign-born populations with TB. Grey literature and reference lists were hand-searched. Risk factor studies were selected for inclusion if they consisted of at least 95% foreign-born populations.

Results

Of 1761 studies identified in the search, 20 were included in the risk factor review. Undocumented immigration status, older age, and social risk factors were consistently correlated with nonadherence; gender, ethnicity, immigration time, education level, adverse side effects, and HIV status were inconsistently correlated; and behavioural risk factors and marital status were consistently not correlated.

Conclusions

This review emphasizes documentation status as a risk factor candidate for further investigation.

Introduction

Today, one-third of the human population is estimated to be infected with latent TB; in 2012 there were 8.6 million incident cases and 1.3 million deaths from the active disease.¹ In low-prevalence countries (defined by WHO as having <20 cases per 100 000 persons) in particular, a rising TB burden in immigrant populations has caused stagnation in decreasing prevalence trends.² In many European countries over half of all TB cases are observed in foreign-born persons, while in the US, the percentage of TB cases attributed to immigrants has increased from 29% in 1992 to 63% in 2012.³ In these low-prevalence countries, TB prevalence in foreign-born populations can be up to 30 times greater than that in the native population, matching rates in high-prevalence countries.² A thorough understanding of the TB treatment dynamics in this group is thus critical for disease control.

Treatment and adherence

Treatment regimens for TB typically last between 6 and 9 months. For active TB WHO recommends 2 months of a four-drug regimen (isoniazid, rifampicin, pyrazinamide, ethambutol) in the initial phase, followed by a continuation phase of isoniazid and rifampicin, lasting 4 to 7 months. Although daily treatment for both phases is recommended, there are also treatment options that include thrice-weekly doses. For latent TB, treatment usually consists of only isoniazid for 6 or 9 months, taken either daily or twice weekly.^4,5 Adherence to this regimen is critical, not only for curing TB in infected populations, but especially for stalling the rise in cases of drug-resistant TB and multidrug-resistant TB (MDR-TB). Improper or incomplete use of medication can drive selection for strains of bacteria that do not respond to standard treatment; as a result, treatment of MDR-TB can take over 2 years and requires more expensive and more toxic drugs. Treatment success rates remain under 50% and mortality rates high: of 450 000 global incident cases of MDR-TB in 2012, there were 170 000 deaths.⁶

Why this review is needed

Three reviews have addressed qualitative evidence surrounding immigrant experiences with TB.^7–9 Although all three found immigration-related factors to be important to TB experience and treatment, none of them engaged with quantitative evidence around that hypothesis. This study systematically reviews quantitative evidence of risk factors for TB treatment nonadherence specifically in foreign-born populations. Studies of risk factors attempt to gather evidence about characteristics of patients who do not adhere to treatment, with the intention of predicting or determining potential causes of nonadherence. Because those risk factors vary widely by context, disease, population, and study type,^10,11 we chose to review evidence for an important population with a shared, though heterogeneous, experience of migration. Specifically, we are interested in knowing what particular aspects of that experience are most correlated with nonadherence. These risk factors can provide a quantitative basis for understanding which populations are most at-risk, facilitating characterization and, as necessary, intervention in those populations. These risk factors can also suggest elements of context that are most closely correlated with outcomes, and identify potential causal risk factors for further analysis.

Methods

Search strategy

Search criteria for initial identification of studies

The search strategy was divided into three criteria that were combined using the ‘AND’ operator: TB; terms that isolated foreign-born populations; and terms that isolated perspectives on adherence.

Syntax and headings adjusted for database usage were used to search for each inclusion criterion. For foreign-born populations and outcomes, search strategies from previous reviews^12–16 were noted and combined into the final search. MEDLINE, Embase, PsycINFO, LILACS, CINAHL, ProQuest (dissertations and theses), and the Social Sciences Citation Index were searched. All search strings, along with number of hits by string, are documented in Supplementary File 1. Grey literature and reference sections from included studies were hand-searched.

Selection of studies

After exclusion of studies that did not address one or more of those terms, studies containing analyses of risk factors for TB treatment outcomes were identified, using the following criteria: at least 95% foreign-born population, or a subgroup analysis with that population; analysis of at least one risk factor comparing adherence between risk factor groups or measurements.

Study populations were required to consist of at least 95% immigrants, with immigrants being defined as: participants whose country of residence is different from their country of origin, regardless of legal documentation or specific location; or participants for whom border changes impacted nationality (through the course of conflict or border negotiation). If a subgroup analysis of immigrants was included as part of a larger study, data for that subgroup were included in this review. Many studies included immigrant status itself as a risk factor; these data were not included, as this review attempted to differentiate risk factors within foreign-born populations. Data about the differences in risk between different immigration statuses (such as undocumented status, refugee status, asylee), were included.

Studies needed to include analysis of at least one risk factor: if this risk factor was measured dichotomously or categorically, then adherence outcomes must have been presented for one immigrant group with the risk factor and one group without (as the ‘control’ group). If risk factors were measured continuously, then the average measurements for adherent and nonadherent groups were required. As the purpose of this study was to examine risk factor characteristics rather than methods of addressing those risk factors, interventions were not included as risk factors. Adherence to medication was defined differently between studies; these various measurements were not used as an exclusion criterion but rather as a basis for discussion.

Data extraction and management

A copy of the data extraction form is presented in Supplementary File 2. Recruitment and measurement methods in particular were carefully noted for discussion. Subgroup analyses were planned for studies with immigrants moving between high-prevalence countries as well as those studying treatment of latent TB. A critical appraisal form for cohort and cross-sectional studies was created, as preliminary screening suggested that these would be the most common type of studies identified; questions were adapted from the Critical Appraisal Skills Programme (CASP) cohort study forms.¹⁷ Critical appraisal occurred in conjunction with data extraction.

Results

Results of the search

Of the 1761 records surveyed, 217 were screened by abstract and 44 isolated for full-length text reviews. Of the 41 texts that could be accessed, 20 were included in this review (Figure 1). A total of 21 studies reviewed for inclusion were excluded.

Figure 1.

Search strategy and study selection for risk factor analysis.

Excluded studies

In total, three full-length texts could not be accessed, and 21 studies were excluded after full-text review. Titles and explanations for exclusion of these studies are presented in Supplementary File 3.

Description of studies

Characteristics of included studies are detailed in Table 1. All attributes are detailed for foreign-born populations only; in cases where demographic data were not detailed or only detailed for a combined population of foreign and native-born patients, they are listed as ‘NR’ (not reported). The majority of studies (12) were retrospective analyses of program or clinic records. These studies were separated by their potential to suggest temporal relationships between outcomes and risk factors. Studies that collected risk factor and outcome data for the same (or unspecified) time point were classified as cross-sectional. If treatment outcomes were recorded after collection of risk factor data for a cohort, the studies were classified as retrospective cohort studies. Only one study was identified as a retrospective cohort study.¹⁸ There were eight prospective cohort studies, which recorded longitudinal data on outcomes for participants entering within specific collection periods. There were no case-control studies, as studies included participants by entry time rather than outcomes. All studies were between-individual comparisons, in that they compared separate groups with and without risk factors rather than outcomes within individuals over time.¹⁹

Table 1.

Characteristics of studies examining risk factors for nonadherence

Ref.	Study design	Location	TB type	Total n	Age	Data analysis	Outcome description
Ailinger & Dear24	Cross-sectional; collected demographic data, interview data, adherence from clinic records.	Immigrants from Latin America in US	Latent	65	29.02	Pearson correlation coefficient for continuous, t-test for dichotomous variables.	Continuous. Sum points for each clinic appointment kept (6) + reported adherence each time (6).
Ailinger et al.32	Cross-sectional; collected adherence data from clinic records and demographics from questionnaire.	Immigrants from Mexico in US	Latent	53	27.3	Pearson correlation coefficient for continuous, ANOVAS for categorical, t-test for dichotomous variables.	Continous. Number of months adhering to clinic appointments.
Ailinger et al.25	Cross-sectional; randomly chose clinic records for adherence and demographic data.	Immigrants from Latin America in US	Latent	160	26.17	Multiple regression for continuous, t-test for dichotomous, ANOVAS for categorical variables.	Continuous. Number of INH doses taken as reported by client and months on treatment (adherence = 8 months therapy).
Arai18	Retrospective cohort study; identified study group and followed through records to treatment outcomes.	Immigrants from Asia in US	Latent	3576	31.23	χ2 for bivariate analysis, logistic regression with 3 variables.	Dichotomous. Excluded died, moved, physician discontinuing. Subjects considered adherent or nonadherent based on final clinical evaluation.
Brassard et al.20	Cross-sectional; reviewed records.	Immigrants from many locations in Canada	Latent	375	14-18	χ2 for bivariate analysis. Multivariable regression by backward-selection, p<0.2 as cutoff.	Dichotomous. Adherence assessed through pill counts, self-report, verification with pharmacies, general attendance. Adequate if patient took >80% total doses within 43 weeks.
Cegolon et al.34	Cross-sectional; reviewed cases in London TB Register.	Immigrants from many locations in London	Active	10 793	NR	Multivariable logistic regression with backwards selection, p<0.05 cut off. Retained some non-significant variables if suspected confounder. models fitted for 2-way interaction, stepwise procedure to final model.	Dichotomous. Failure: TB caused/contributed to death or relationship unknown; treatment outcome not reported; lost to follow-up; stopped treatment deliberately (not for side effects); unknown.
Codecasa et al.26	Prospective cohort; risk factor data collected prospectively and outcomes recorded in treatment.	Immigrants from many locations in Milan	Latent	8586	27.6	χ2 for categorical; t-test for continuous variables. Multivariable logistic regression with variables significant in bivariate analysis.	Dichotomous. Non-completion if discontinued for any reason (before 6 months).
Coly & Morisky21	Prospective cohort; combined with experimental study and participants followed over time.	Immigrants from many locations in US	Latent	610	15	Pearson’s χ2 for categorical, t-tests for continuous variables; Multivariable logistic regression with condition forward procedure.	Dichotomous. Non-completion of treatment measured by discharge summary according to patient’s chart.
Helbling et al.35	Prospective cohort; patients identified and physician questionnaires used to collect outcomes after 10 months.	Immigrants from many locations in Switzerland	Active	152	30	Logistic regression for each risk factor individually. Main effects model including all predictors estimated.	Categorical. Potentially unsatisfactory: Failure, default or transfer out.
Jiménez-Fuentes et al.30	Prospective cohort; combined with experimental study and participants followed over time.	Immigrants from many locations in Spain	Latent	590	26	χ2 test with Yates correction.	Dichotomous. Adherence was defined as >80% adherence based on urine testing + clinic appointments kept.
Kapella et al.36	Prospective cohort; risk factor data collected prospectively and outcomes recorded for treatment.	Immigrants from many locations in Thailand	Active	791	33	Multivariable analysis with log-binomial regression model. Chosen for inclusion if p<0.2 in bivariate analysis or if literature suggest association.	Categorical. WHO guidelines for default (death, transfer, change, continuing treatment) compared with cured/completed/failed.
Lake et al.37	Cross-sectional; risk factor and outcome information collected simultaneously from database.	Immigrants from many locations in UK	Active	21 954	NR	Separate multivariable regression models for age <45 and >44 years.	Dichotomous. Completion based on clinical information after 12 months of start.
Minetti et al.27	Cross-sectional; retrospective reviews of medical files.	Immigrants from Burma in Thailand	Active	2425	35	χ2 for categorical, t-tests for continuous variables. Multivariable model with Cox regression.	Dichotomous. Treatment success defined as cured (Smear+ to Smear-) or treatment completed (did not fail treatment, die, default, transfer).
Minodier et al.22	Prospective cohort; risk factor data collected prospectively and outcomes recorded for treatment.	Immigrants from many locations in Canada	Latent	545	12.17	Binary regression, multivariable regression with variables significant or ‘close’ to significant.	Dichotomous. Adherence defined as >80% doses in <120% time planned.
Mitruka et al.28	Cross-sectional; retrospective reviews of National TB surveillance system.	Immigrants from many locations in US	NR	7098	NR	Multivariable models excluded non-significant predictor variables, excluded missing/unknown, assessed collinearity through model convergence.	Dichotomous. Failure to complete treatment <12 months.
Nelson et al.23	Cross-sectional; merged TB case data with another demographic dataset (temporality not explicit).	Immigrants from Tibet in India	Active	466	NR	Multivariable regression with p<0.15 risk factors, variables assessed for collinearity.	Categorical. Defaulted based on clinic data (other outcomes reported, but default used for this review).
Nissapatorn et al.38	Cross-sectional; single data collection from medical records.	Immigrants from many locations in Malaysia	Active (PTB)	263	33.3	χ2 or fisher’s exact test.	Categorical. Default (2+ months consecutive treatment interruption), transferring out (transferred to another unit, treatment outcome unknown) or <9 months therapy.
Orr et al.39	Cross-sectional; collected all data from medical records.	Immigrants from many locations in Canada	Latent	523	NR	χ2 test for significance.	Dichotomous. Failure to initially appear at clinic or attend follow-up visits.
Scotto et al.33	Prospective cohort; physicians completed one risk factor questionnaire and one at the end of treatment.	Immigrants from many locations in Italy	Active	300	34	χ2 test for significance.	Dichotomous. Based on physician questionnaires about compliance.
Trauer & Krause29	Prospective cohort; risk factors collected and completion data sought 9 months later.	Immigrants from many locations in Australia	Latent	93	NR (15 for screening)	Multivariable logistic regression.	Dichotomous. Completing >80% doses of therapy in <12 months.

INH: isonicotinylhydrazine or isoniazid; NR: not reported; OR: odds ratio; PTB: pulmonary TB; RR: relative risk.

Location

The majority of papers (14) did not specify a country of origin, though all destination countries were reported. This discrepancy was largely a result of study type; because many studies used data from clinics or programs in the destination country, immigrant populations from different origin countries were included within the same analysis. There were five studies characterizing immigrants moving between countries with high TB prevalence; the remainder of studies described immigrants in countries with low TB prevalence.

Participants

Three studies examined adherence in children or adolescents specifically^20–22; of the remaining 17 studies, 16 (with the exception of Nelson et al.²³) had mean ages below 35. Only two studies reported specific sampling procedures; one included a ‘convenience’ sample of patients ‘available for interview at the time of the study’ (p. 21)²⁴ and another included a random sample of patients on record at the clinic, with randomization not reported.²⁵ All other studies examined all patients that entered a clinic or TB program within a specific time range. In total, six studies focused on patients within a particular clinic, while the remaining 14 studies examined all records within a particular location (city, county or country-wide TB program).

Treatment type

It is useful to note that some studies contained participants on different treatment regimens, especially in cases where studies examined clinic or program records spanning decades of treatment.^26–28 The most common treatments were isoniazid for nine months for latent TB and isoniazid, rifampicin, ethambutol and pyrazinamide for two months, followed by isoniazid and rifampicin for four months in cases of active TB. Overall adherence, though not always reported, ranged from 49.5%²⁹ to 84.5%.²⁸

Risk factor measurements

For retrospective studies, risk factors were identified from clinic or program reports (with one exception being Nelson et al.²³). In prospective cohort analyses, studies used a combination of records collected during patient treatment^25,27,30,31 and active collection of data separate from clinical records. This separate collection of data allowed for targeting of specific risk factor data from all participants (mitigating the problem of missing clinic data), but opened avenues for patient dropout, or refusal to participate in the study despite being enrolled in the TB control program. None of the studies, however, cite significant rates of dropout.

Outcome definitions

Definitions of adherence varied significantly. Three studies monitored it continuously, characterizing number of months adhering to treatment for each patient,^24,25,32 while the remaining 17 studies characterized it as a dichotomous or categorical variable. Most broadly, measurements were divided by study type: while five of nine prospective cohort studies directly reported process-based measures, all cross-sectional and retrospective studies used outcome-based measures. The five studies using process-based measures included clinic appointments kept,^24,25,30,32 urine testing³⁰ and physician questionnaires.³³ Outcome-based measures included WHO guidelines for cure and treatment completion,⁶ 80% completion as a cut off for adherence, and ‘noncompletion’ based on a number of poor outcomes (such as death, treatment failure, stopping treatment).

Data analysis

Of 20 studies, 14 used multivariable regression models to characterize independent associations of risk factors, most often adjusting for age, gender and region of origin. In four studies, all individually-examined variables were included in the regression model,^25,26,29,40 while in six studies only variables meeting a cut off for significance in bivariate analysis were included. Lake et al.³⁷ separated groups into analyses based on identified confounders. Only two studies explicitly tested for collinearity,^23,28 though no results of this analysis were mentioned in results.

Methodological critique of studies

The results of critical appraisal are in Table 2.

Table 2.

Critical appraisal of studies examining risk factors for nonadherence

	Recruitment acceptable	Risk factor measures appropriate	Outcome measures appropriate	Follow up complete	Follow up long enough	Confounding factors accounted for
Ailinger & Dear24	N	Y	Y	Y	M	None
Ailinger et al.32	Y	Y	M	Y	Y	None
Ailinger et al.25	Y	Y	M	Y	M	Age, education, years migrant, side effects, public health nurse
Arai18	Y	Y	M	Y	Y	Country of origin, speaking English, years in US
Brassard et al.20	Y	Y	M	N	Y	Age, gender, region of origin, time in country, chest radiograph status, treatment adverse effects, number of family members screened
Cegolon et al.34	Y	Y	Y	Y	M	Gender, age, previous treatment, ethnic group, index of multiple deprivation score, chest clinic caseload, site of disease
Codecasa et al.26	Y	Y	Y	Y	Y	Sex, age, reason for treating, HIV status, adverse effects
Coly & Morisky21	Y	Y	Y	Y	M	Age, ethnicity, family status, knows therapy prevents disease, clinic waiting time, language spoken, sexual activity, gang membership, history of incarceration, medication taking behaviour score
Helbling et al.35	Y	Y	Y	Y	Y	Age, sex, origin
Jiménez-Fuentes et al.30	Y	Y	Y	N	Y	None
Kapella et al.36	Y	Y	Y	Y	Y	Sex, age, marital status, location, mobility, previous treatment, HIV, diabetes, cough, incarceration, TB type, facility, category, MDR
Lake et al.37	M	Y	Y	M	Y	Sex, age, ethnic group, place of birth, TB history, disease site, INH susceptibility, distance to TB treatment centre, TB centre load
Minetti et al.27	Y	Y	Y	M	Y	Female, age <15 years, refugee status, smear-positive pulmonary TB
Minodier et al.22	Y	Y	Y	N	Y	Origin, planned treatment duration, INH pills vs syrup, age, delay presence relatives
Mitruka et al.28	Y	Y	Y	M	Y	Sex, race, age, disease site, HIV, previous diagnosis, smear, substance, homeless, correctional facility, DOT, provider
Nelson et al.23	Y	M	M	M	M	Age, region, gender, default, previous TB
Nissapatorn et al.38	Y	Y	N	M	M	None
Orr et al.39	Y	Y	M	M	Y	None
Scotto et al.	Y	Y	M	Y	Y	None
Trauer & Krause29	Y	Y	Y	Y	Y	Region, sex, age

INH: isonicotinylhydrazine or isoniazid; M: maybe or not reported; MDR: multidrug resistance; N: no; Y: yes.

Participant selection

The use of clinical records in most of these studies confined analysis to only those patients who initiated treatment. This initiation can occur by treatment seeking or, in the case of latent TB, screening programs. Many countries require latent TB testing as part of immigration procedure, but do not always require treatment.⁴¹ The adherence risk factors measured in these studies, then, are not applicable to the immigrant population as a whole but rather those initiating treatment after being pre-screened for latent TB based on recommendation or eligibility. For example, Brassard et al.²⁰ noted that of 542 students with positive tuberculin skin test results, only 484 presented at a clinic and of those, 377 were started on therapy. As only the populations undergoing treatment may be targeted for interventions aiming to increase adherence, this selection does not change applicability of results; nevertheless, it is useful to consider that it has occurred.

Missing data

Though studies were consistent in reporting missing data in bivariate analysis or exclusion of records, only Cegolon et al.³⁴ and Arai¹⁸ described how it affected multivariable analyses. In particular, Arai found that ‘the collection of data on the covariates, such as HIV status, homelessness…and long-term care facility status, has not been required for those receiving treatment for latent tuberculosis infection (LTBI),’ so missing data for these variables precluded their examination in analysis. Cegolon et al. found that of 12 924 initial observations only 6253 were available for multivariable regression analysis because others were missing data. In the reporting of this missing data, while only two records failed to report age and one gender, 2157 records were missing time since UK entry.

Outcome measurements

A few studies did not note which outcomes were aggregated, only defining groups in categories such as ‘suboptimal,’ ‘no improvement’ or ‘compliant’.^23,33,39 It is useful to note the variation in these measures and consider the degree to which results are directly related to adherence versus other factors in poor outcomes. For example, in the case of using death as an outcome for adherence, if the incidence of drug-resistant TB (which can cause death even with full adherence) is higher in the risk factor or control group, then analyses based on this outcome might not reflect nonadherence risk. Generalized cut off times may also not reflect the reality of treatment requirements: for example, Minetti et al. found that ‘failure to timely completion of treatment’ was more related to prolonged treatment rather than failure to complete treatment at all.²⁷ Thus, some categories of outcomes used did not necessarily isolate on adherence behaviours.

Data analysis

The significant heterogeneity in multivariable models constructed, as well as the methodologies used to construct those models, inhibits comparison of risk factors across studies. As this synthesis seeks to interpret independent correlations of different risk factors with outcomes, the variation in variables accounted for can be misleading. Omission of confounding variables from multivariable models can result in misleading correlations of ‘proxy’ risk factors with outcomes; on the other hand, addition of collinear or overlapping variables can obscure correlations between risk factors and outcomes.^19,42

Risk factors for nonadherence

Demographic variables

Codecasa et al.²⁶ and Minetti et al.²⁷ both found males to be at higher risk for noncompletion (OR 1.42, 95% CI 1.29, 1.56 and aOR 1.33, 95% CI 1.09, 1.62 respectively), though of the 13 studies examining gender, the remaining 11 found no significant correlation with adherence (Figure 2). Increasing age as a continuous variable was found to be associated with less adherence (p=0.003, η=0.002)¹⁸; this effect was also seen in studies assessing older age as a dichotomous variable.^28,36,39 Three studies found children (<15 years) to be more likely to complete treatment.^27–29 Nelson et al.,²³ in contrast, found no association (aOR 1.2, 95% CI 0.5, 2.7) between age and outcomes. They studied, however, a population of older Tibetan-born refugees in India, for whom the comparison group of younger patients was defined as those below age 50 (155/460 people). It is possible that correlations with age are only apparent in younger cohorts.

Figure 2.

Forest plot of studies with male gender as a risk factor for nonadherence. Ailinger and Dear²⁴ and Ailinger et al.³² not depicted due to use of continuous outcomes.

Of seven studies examining the correlation between ethnicity or country of origin and adherence, only two note statistically significant effects. Arai¹⁸ found that Chinese immigrants in San Francisco were less likely to be nonadherent than Filipino immigrants after controlling for years in the US and English-speaking ability (aOR 0.75, 95% CI 0.60, 0.94); Trauer et al.²⁹ found Southeast Asian immigrants to be more nonadherent than African immigrants (OR 6.21, 95% CI 1.59, 24.30) though their sample of Southeast Asian immigrants was small (n=15) and covariates were uncontrolled for. Several studies did note, however, that ethnicity was significantly related to age for certain populations and migration patterns.^18,23,40 For example, Nelson et al.²³ found that in a refugee camp in India, Tibetan-born refugees were significantly older, less educated, and spent less time away from the camp compared to populations born in India. Helbling et al.,³⁵ who did not include ethnicity as a risk factor, noted that ‘geographic origin’ was highly correlated with legal status, age, and gender. Thus, differences in adherence based on ethnicity may be accounted for by resulting differences in migration background.

Immigration-related variables

Undocumented or unknown immigration status was found to be significantly associated with treatment noncompletion in all five studies that examined it as a risk factor (Figure 3).^{26,27,30,35,40} Minetti et al. noted that in Thailand, undocumented Burmese migrants, ‘if arrested, were sent back to Myanmar where tracing was not possible’ (p. 1593)²⁷ and that this loss was a significant source of defaulting. Helbling et al. found, similarly, that ‘all 16 [immigrants of unknown/other documentation status]’ had ‘transfer out [outcomes], two known to be involuntary’ (p. 52).³⁵ Thus, loss to follow-up after repatriation was suggested to contribute to poor outcomes noted for undocumented populations. Helbling et al.³⁵ found a significant difference in adherence between patients with unknown/other immigration statuses as compared to asylum seekers/refugees and foreign workers. There was also a significant difference in the proportion of ‘legal’ statuses between ethnic groups in Jiménez-Fuentes et al.’s study (χ²=17.563, df=3, p<0.05)³⁰; no other studies explicitly examined correlations between immigration status and other risk factors.

Figure 3.

Forest plot of studies with immigration status as a risk factor for nonadherence.

Risk factors related to immigration time or age, however, had less consistent correlations. Total length of time spent in the destination country was inconsistently associated with risk, with Cegolon et al.³⁴ finding recent entry (<2 years) to be a greater predictor of failure to complete treatment than later entry (OR 2.26, 95% CI 1.63, 3.12 vs OR 1.53, 95% CI 1.10, 2.12, as compared to UK populations) while Arai¹⁸ found the opposite correlation, with recent entry (<5 years) predicting decreased nonadherence (OR 0.964, 95% CI 0.953, 0.974). Four other studies^20,23,30,32 found no significant effect of time since arrival. Arai also found age at immigration not to be significant in predicting treatment outcomes (p=0.108). Two studies that examined English-speaking ability (in the US) explicitly suggested it to be a ‘proxy for acculturation’ or cultural ‘assimilation;’ they found, however, that immigrants who spoke more English were at increased risk of not completing treatment.^18,21 One study in adults found English-speaking populations to be at higher risk of nonadherence after adjusting for years in the US and country of origin (OR 2.40, 95% CI 1.91, 3.00)¹⁸; the other study²¹ found children speaking primarily English at home more likely to be adherent in multivariable analysis, but this association was not significant at the 95% CI (aOR for adherence 0.39, 95% CI 0.15, 1.03). Overall, there is no consistent correlation between immigration time or age, and treatment adherence.

Social risk variables

Social risk variables were generally found to predict treatment failure, though sample sizes for many of these variables were small (Figure 4). Mitruka et al.²⁸ found that being in a correctional facility at time of diagnosis was associated with increased noncompletion rates after adjusting for variables including age, gender, and substance abuse (aOR 2.76, 95% CI 1.99, 3.81); the same was found by Coly and Morisky²¹ for any history of incarceration (OR 2.01, 95% CI 1.06, 3.82), though this effect was nonsignificant in multivariable analysis when adjusted for factors including age, ethnicity, and gang membership. Kapella et al. did not find a history of incarceration to be correlated with nonadherence, though their sample size was small (n=11).³⁶ Homelessness in the past year was found to predict failures (aOR 2.05, 95% CI 1.46, 2.88),²⁸ and patients identified through shelters or drug treatment facilities were also more likely to discontinue treatment (OR 2.87, 95% CI 2.55, 3.22).²⁶ Coly and Morisky²¹ also found that being a gang member was associated with increased nonadherence in adolescents (OR 3.91, 95% CI 1.17, 13.06), though the effect was found in a very small risk factor group (n=11) and was nonsignificant in multivariable analysis. Interestingly, behavioural risk variables including smoking,³⁰ alcohol,³⁰ and drug use^28,30,36 were not significantly correlated with noncompletion.

Figure 4.

Forest plot of studies with social risk factors for nonadherence.

There was disagreement about the degree to which education and employment were associated with adherence. Jiménez-Fuentes et al.³⁰ found that those with a low level of education were more likely to default from treatment (OR 1.73, 95% CI [1.07, 2.80]), while three other studies^23,24,32 failed to find significant associations between education and adherence. Comparison was difficult, because two studies^24,32 used education and adherence as continuous variables; thus, it is possible that even if increased education is not associated with adherence, having significantly low levels of education could be. Nelson et al.²³ conducted their study in Tibetan refugees in India, where overall education levels were lower; perhaps having less education in that context, then, was not significantly associated with adherence. Jiménez-Fuentes et al. did not control for association of employment and education with variables such as immigration status or age, even though over half the participants in their younger cohort (mean age 26.1) had undocumented immigration statuses. Kapella et al.²³ and Nelson et al.²³ found no correlation between adherence and specific types of employment in multivariable analysis (aOR NR and aOR 1.0, 95% CI 0.3, 3.0, respectively), though Jiménez-Fuentes et al.³⁰ noted an association between unemployment and increased default (OR 1.91, 95% CI 1.30, 2.79).

Relationship-related factors were also inconsistently correlated with adherence. Presence of both parents was a predictor of less noncompletion in children (aOR 0.57, 95% CI 0.34, 0.98),²¹ while not living with family, in adults, was a predictor of nonadherence (OR 3.7, 95% CI 2.54, 5.4).³⁰ Conversely, the presence of relatives >18 years old was found to be a predictor of nonadherence (aOR 1.56, 95% CI 1.00, 2.43), though authors attributed the effect to ‘collection of socio-demographic features rather than psychosocial ones’ (Minodier et al., p. 73).²² Marital status had an insignificant correlation with adherence in two studies^23,36 as did self-report of a significant other’s help,²⁴ though the sample of patients reporting help at all was small (n=18, t(65)=−0.33, p=0.741).

Access-related variables

In two studies, variables related to the degree of interaction with health care services were found to be correlated with adherence outcomes. Delay in tuberculin skin test presentation after an initial screening of students was a predictor for poor adherence (aOR 1.6, 95% CI 1.12, 2.28).²² It is unclear whether this delay was a product of patient or system delay; the study, interestingly, framed this delay as a possible ‘delay in management [that] could lead parents to doubt the seriousness of a positive LTBI [test]’ (p. 73). Lake et al.,³⁷ similarly, examined the effect of treatment centre centralization in the UK, and found that traveling more than 7.3 km to TB treatment centres correlated with increased rates of default in both younger (aOR 1.23, 95% CI 1.10, 1.39) and older (aOR 1.41, 95% CI 1.17, 1.90) patients.

Disease-related variables

The impact of medication side effects on treatment adherence is mixed. Studies using adherence as a continuous variable found either no correlation with adverse effects (t(65)=−1.45, p=0.15),²⁴ or correlations that were not significant in multivariable analysis.²⁵ One study in adolescents also found no significant correlation with adherence (aOR 0.83, 95% CI 0.31, 2.26), but it is useful to note that this study only observed adverse effects in 5.8% of patients (n=22).²⁰ One larger study found adverse side effects to have a significant impact on treatment non-completion (OR 1.33, 95% CI 1.15, 1.53, n=8586); unlike other studies, however, a significant proportion of that non-completion included interruption of treatment prescription by the leading physician due to adverse effects.²⁶ In discussing the unexpected lack of adverse effects in certain immigrant populations, Codecasa et al. suggested ‘language/cultural barriers’ as one possible explanation for ‘reduced reporting of adverse effects’ (p. 906).²⁶ Hence, language barriers could pose a threat to validity of risk factor measurements.

There was also mixed evidence suggesting HIV status as a risk factor for nonadherence. Mitruka et al. noted an increased risk of nonadherence in HIV+ patients that seemed to be manifested in greater numbers of loss to follow-up (aOR 2.10, 95% CI 1.58, 2.87)²⁸; Kapella et al. found a slightly significant correlation with adherence (OR 1.6, 95% CI 1.0, 2.4) that was not present after multivariable analysis with variables including age, mobility, and DOT.³⁶ It is interesting but unsurprising that in a cohort of patients on treatment for latent as opposed to active TB, Codecasa et al.²⁶ found very low incidence of HIV (46/8586). They also found no correlation between HIV status and treatment adherence (OR 1.05, 95% CI 0.55, 1.99).

Subgroup analyses

High-prevalence countries

A subgroup analysis was planned using the subset of studies from destination countries with high prevalence of TB. Four studies taking place in Thailand,^27,36 Malaysia,³⁸ and India²³ were identified. All four studies were of patients undergoing treatment for active TB. Though no strong differences were noted between results of these studies as compared to the full cohort, all four of the studies comment on the effect of migratory lifestyles on medication follow-up.

Latent vs active TB

No consistent differences were noted between studies or results of studies examining treatment for latent vs active TB.

Discussion

This study is the only systematic review of risk factors for nonadherence to completion of TB treatment. The studies in this review are largely taken from TB control programs in clinics, jurisdictions, or countries. Only undocumented status, older age, and social risk factors (homelessness, incarceration) consistently correlated with nonadherence. Gender, ethnicity, HIV status, adverse side effects, and education had varying correlations with negative outcomes based on context, while behavioural risk factors and marital status were consistently not associated with adherence. Though no subgroup differences were observed for high-prevalence countries or latent versus active TB, all studies of high-prevalence countries observed the association between migration and adherence.

Comparison to existing literature

The results of this review align with those from HIV and other medication adherence reviews. Social support, age, and gender were all found to have inconsistent correlations with nonadherence risk.^10,43 This review also concurs with previous studies that found that exogenous risk factors were more predictive of adherence than endogenous factors of age or gender.¹⁰ In our review, immigration status and social risk factors were found to be the strongest predictors of adherence. Although exogenous risk factors were examined in fewer studies, both direction and significance of effect were more consistent than those for permanent risk factors. This finding is encouraging, because it leaves the possibility of intervention open, in the case that these changeable risk factors are shown to be causal in future work.

Implications for future research

This review underscores the importance of having appropriate outcome measures, and conducting analyses that account for confounding effects in relationship to the variable of interest. Notably, several studies found that the potential risk factor of ethnicity was significantly related to other risk factors, including age and immigration status, which could obscure understanding of how ethnicity may relate to TB adherence. A more nuanced understanding, however, is especially important when attempting to relate risk factors to qualitative data – for example, data examining cultural differences towards western treatments – or intervention design. Additionally, more information on immigration status as a risk factor would be interesting, as would relationships between immigration status, different host countries’ approaches towards health care of undocumented immigrants, and medication adherence.

Implications for policy and practice

The most significant finding related to policy and practice is perhaps the consistent association this review noted between immigration status and treatment adherence. Although the largely cross-sectional studies here cannot prove immigration status has a causal relationship with treatment adherence, this relationship is nevertheless significant, and its nature important to understand on a local level. It may be useful to think critically about how individual countries’ immigration policies may affect patients’ willingness to both access and maintain treatment for TB. Additionally, the association between societal risk factors (correctional facilities, homelessness, gang membership) and nonadherence could be used to better target or design interventions specific to patients in these groups.

Strengths and limitations of the study

This study’s systematic search and selection criteria ensured the synthesis of extensive journal-level evidence on this topic, as well as the auditability of the inclusion process. Studies included in the review come from a variety of countries and treatment programs, strengthening generalizability for risk factors examined. Additionally, although meta-analysis was not performed, data for each risk factor were rigorously examined and compared with observations of methodological heterogeneity. This comparison allowed for more nuanced interpretation and auditability of results.

The quality of studies examined in this review varied significantly, especially with regards to adjustments for confounding variables and reporting of missing data. Only two of the studies reported specific sampling procedures, which is critical in risk factor inferences from cross-sectional studies. It was especially difficult to ascertain impacts of individual risk factors examined in studies that only used bivariate analysis. Additionally, reporting of risk factors was variable, both with regards to risk factor measurement and reporting of missing data for risk factors. On an aggregative level, there was significant heterogeneity in outcome measurements, risk factor measurements and variables accounted for in regression models. This heterogeneity is widely-cited as a source of difficulty in risk factor review; DiMatteo in particular notes that the lack of a gold standard in adherence studies impedes meta-analysis and comparisons.¹⁰ Efforts were made to intentionally note and suggest sources of heterogeneity in the review of each risk factor group. Our limitation of studies to those published in English may also have omitted some of the relevant research, especially from high-prevalence countries.

Conclusion

The results of this review emphasize the correlation between adherence and age, immigration status, and social risk factors, as well as the inconsistent association of other variables, including ethnicity. Notably, unknown or undocumented immigration status was the risk factor most consistently correlated with nonadherence. We suggest that this may be a worthy avenue of investigation for future studies, especially as this population is one that may interact with multiple countries – both the country of destination and, in the case of deportation, the country of origin. It would be useful to examine the causality of this risk factor (i.e., does the experience of being undocumented causally affect patient adherence to treatment and, if so, how), as well as the possibility of interventions supporting greater adherence in this population.