Introduction

Tuberculosis (TB) is still today a major global public health problem due to its high impact in terms of mortality and morbidity, particularly in economically active groups of low- and middle-income countries.

Despite being a disease with an effective and affordable therapy, treatment success rates are disappointingly low [1] in many settings. A primary cause of low success is due to poor adherence to the challenging treatment regimen that imposes an important burden on patients [2–3]. Several strategies have been proposed to improve TB treatment adherence using financial incentives, such as conditional cash transfers (CCTs). CCTs can offer a positive incentive conditional on a certain behavior, such as an action focused on improving a health outcome [4]. In many countries, CCT programs form the backbone of social security policy as a form of social assistance to improve uptake of health interventions. For TB treatment, CCTs support individuals contingent on taking treatment and attending follow-up appointments. Such a policy offers the advantage of assisting an individual during a critical time when they are required to abstain from work or other activities that could increase the risk of disease transmission to others. Although CCTs are recognized as a potentially powerful tool to promote healthy behaviors, the formal evaluation of the impact of these strategies has been very limited in TB control, particularly in Latin America and the Caribbean. It is important to understand if such programs are effective since they generate costs of implementation and monitoring.

In 1986, the province of Buenos Aires, Argentina, which concentrates 48% of the more than 10,000 notified cases per year in the country, passed a law addressing the use of a CCT as a social support policy to promote adherence to TB treatment for vulnerable patients. To date, only one study reported that patients receiving this financial incentive had more adherence and a higher success rate of treatment [5]. Although an important finding, a next step needed is to consider and assess other potential determinants of TB treatment outcomes, such as access to healthcare, primary source of care, community outreach programs, treatment modality received, comorbidities, and several socioeconomic characteristics. By including and controlling for these potentially confounding variables, we can ascertain a clearer picture of the specific effect of the CCT program on TB outcomes.

Methods

Results

Recruitment commenced on September 2011, and after reaching the target sample size on June 2014, the last patient follow-up was completed on December 2014. In total, we recruited 962 patients, but in 21, we could not confirm their final group allocation (registered or not to CCT). Thus, of the remaining 941 patients with information on CCT allocation, 377 were registered for the CCT and 564 were not. In each group, there was some information missing on treatment outcome (see Fig 1). Overall, the rate of treatment abandonment was 16.3%. Of the 153 patients who abandoned treatment, 31 did so before completing the second month of treatment, 77 by the fourth month, and 45 by the last visit at 6 months.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart.

https://doi.org/10.1371/journal.pmed.1002788.g001

Table 1 shows the sociodemographic characteristics and risk factors by group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sociodemographic characteristics and risk factors by group.

https://doi.org/10.1371/journal.pmed.1002788.t001

Crude analysis shows that being registered for the program was strongly associated with treatment success and default rates, with those not under the CCT showing significantly lower success rates (69% versus 83%) and higher default rates (20% versus 11%), respectively, both P < 0.001.

As expected, there were several differences among those included and not registered in the CCT program. Registered patients were more commonly not employed or had an informal job, lower income, and lack of health insurance. DOT or mixed treatment modality was more prevalent than SAT as well as having a primary care center as their main source of care. Age, gender, marital status, educational level, smoking, drug use, HIV status, distance to healthcare center, and other variables related to healthcare were not associated with being or not being registered in the program.

In addition to the CCT program, other individual factors significantly associated with higher default rates were nationality, alcohol and drug use, smoking, type of job, lowest income tertile, and lack of health insurance. Younger patients also showed a higher risk of abandonment. Treatment modality was strongly associated with success and default rate, with the SAT group showing significantly higher default rates than those receiving DOT or mixed regimes (20.3% versus 8.3% and 7.5%), respectively.

Regarding system-level variables, receiving care at a hospital versus a primary care center, lack of community outreach programs and lack of periodic supervision from the TB program also were related with higher default rates.

We used a multilevel logistic regression model to estimate the adjusted effect of the program on default and success rates, adjusting for all of the identified potential confounders at the individual and the healthcare system level.

As seen in Table 2, compared to patients not registered in the program (reference category), the crude odds ratio (OR) for abandonment for those registered was 0.45 (95% CI 0.30, 0.68, P < 0.001), suggesting a significantly lower risk in this group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Multilevel model.

Factors associated to incomplete treatment (default).

https://doi.org/10.1371/journal.pmed.1002788.t002

Multilevel models

The variability in default rates was substantial among the different HCFs. The intracluster correlation coefficient (ICC) was significant (ICC 0.123 [95% CI 0.08–0.17]), meaning that although the individual patient characteristics explain most of the variability, approximately 12% of the total variance can be explained by characteristics of the system and HCFs (level 2) (primary clinic- versus hospital-based care, CCT, community outreach, training of healthcare teams, treatment modality prescribed, etc.) The multivariable model in Tables 2 and 3 show that, after adjusting for the most important individual and health system factors, being registered for the CCT was associated with a substantially lower odds of default and higher odds of success: adjusted ORs 0.36 [95% CI 0.23, 0.57], and OR 2.9 [95% CI 2, 4.3] respectively, both P < 0.01. Other variables associated with a higher risk of incomplete treatment were SAT, younger age, lack of insurance, lower income, and use of alcohol and illicit drugs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multilevel model: Factors associated to treatment success.

https://doi.org/10.1371/journal.pmed.1002788.t003

Since this was a prospective cohort study and to facilitate interpretability of main results, we used the adjusted OR obtained from the multilevel logistic regression to estimate the corresponding adjusted relative risks (formula in S2 Text) [17]. Hence, the adjusted relative risk (RR) for successful treatment in patients in the CCT group was 1.25 (1.18, 1.31) and for incomplete treatment was 0.41 (0.27, 0.62), both P < 0.001.

The PS matching yielded a good balance among the variables used for adjustment (see Table 4). The region of common support for PS matching was between 0.15 and 0.72. Individuals in non overlapping areas of the PS were not considered in the matched analysis (49 controls and 31 in CCT original groups).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Comparison of patient characteristics between CCT and control groups before and after PS matching.

https://doi.org/10.1371/journal.pmed.1002788.t004

Using PS matching and IPWRA as alternative methods to adjust for confounders yielded essentially the same results as the multilevel models, with estimated adjusted treatment effects of an absolute reduction of abandonment of −12.6% 95% CI (−7.7%, −17.5%) and an absolute increase in treatment success rate of 15% 95% CI (9%, 21%), both P < 0.001 (S2 Text).

Discussion

Our results show that patients registered for the CCT had greater success rates and were less than half as likely to have incomplete treatment after controlling for individual and healthcare system factors as potential confounders.This suggests that the registration for this financial incentive (i.e., the intent to grant this CCT) had a significant effect on adherence to TB treatment, independently of age, educational and income level, employment and marital status, source of care, availability of community programs, and modality of treatment received. Interestingly, the exposed group could nearly achieve the WHO goal of at least 85% completed treatment [18].

Although TB care is provided free of charge by the public system, patients incur important direct expenses to access the treatment and may lose or reduce their source of income if they cannot work. This program intends to provide social protection and promote treatment completion among TB patients. It is made effective through the payment of a monthly amount, which today represents a percentage of the minimum wage category of the public administration in the province, to all eligible patients identified and incorporated into the Provincial TB Control Program (PTP).

Beneficiaries are mandated to keep health controls, treatments, and other conditions established by the PTP; failure to do so may result in the loss of the benefit. The process needs to be initiated by a health professional. A social worker and a physician evaluate each case, taking into account the severity, the socioeconomic situation, the community risks, and the most susceptible age groups. Once the candidate is individualized, they advise him to complete the application form and the relevant obligations. When this process is completed, the social worker and the attending physician state in the same request that the patient is eligible for the program. This request must be attached to the social report and the medical record and sent to the PTP, who is responsible for evaluating the application and granting or denying the requested subsidy. This is a long, complex, and bureaucratic process that generally lasts more than the duration of the treatment, and at times, the patients receive the benefit up until 1 year after completion. Thus, unfortunately, this “TB-specific” intervention has not been broadly available to patients in the province.

Another systematic review and meta-analysis assessed the effects of social protection on TB treatment outcomes in low- or middle-income and in high-burden countries [19]. This review included 12 studies evaluating financial interventions (within them, only one retrospective cohort evaluating a CCT [20] and one randomized clinical trial (RCT) evaluating a monetary incentive with vouchers) [21]. The other RCTs evaluated food incentives [22,23] and social support or educational interventions [24–29]. This systematic review showed that social protection was associated with TB treatment success, cure of TB patients, and reduction in risk of incomplete treatment (TB treatment default). However, the authors also concluded that overall quality of evidences regarding these effect estimates is low. Therefore, there is limited evidence to support that sustained incentive programs can improve long-term adherence to TB treatment. Most of the evaluated interventions were isolated incentives and not a formal social protection policy program, as is the case in our study, which has existed for more than 20 years but has not yet been widely implemented.

Some observational studies were also conducted. A review of the impact of cash transfer and microfinance interventions for TB control [4] found that these interventions have the potential to improve people’s access to TB services and reduce people’s vulnerability to TB by improving households’ socioeconomic situations. However, given the relatively short follow-up period in many of the studies, little can be concluded concerning the sustainability of these findings. This study concluded that synergies between social protection interventions and TB control programs could be effective. A cross-national statistical modeling analysis performed in 21 European countries [30] showed that an increase in social protection spending was associated with a decrease in the number of TB case notifications, estimated incidence rates, and mortality rates, but no association was found on smear-positive treatment success. Another observational study performed in the city of New York [31] showed that the odds that a patient would adhere to therapy were greater with increased incentives, independently of clinical, demographic, or social factors.

Several other studies with different methodologies provided similar insights supporting the use of cash transfer programs [32–38]. However, important methodological issues in the published studies (small sample size, contextual factors affecting the intervention, interventions conducted in traditionally hard-to-reach or marginal populations, lack of adequate adjustments, etc.) might limit the evaluation of the potential effect of this strategy to reduce abandonment of TB treatment.

One recent RCT [39] evaluated the impact of a social support program and a CCT on prevention and treatment success in shantytowns in Peru. It showed that treatment was successful in 64% (87/135) of patients receiving the socioeconomic support versus 53% (78/147) in the control arm.

As mentioned before, our study is an observational study evaluating an intervention not randomly assigned. Hence, some selection bias may play a role since the two compared groups have differences in certain characteristics associated with the outcomes, and there might be some unmeasured residual confounding not accounted for by this analysis. However, we used rigorous methodologies to control for confounders and for site and population selection. The consistent results of the multilevel logistic regression, the PS approach, and the causal inference analysis used to adjust for multiple factors suggest a significant effect of the CCT program on treatment success and abandonment. Also, the study population was drawn from broad heterogeneous healthcare settings that included semirural and urban settings that likely have similarities to other public healthcare systems in Latin American or other low- and middle-income countries (LMICs).

In 2014, there were 9,600 notifications of new cases in Argentina; of those, almost 8,000 were pulmonary TB and 90% of them older than 15 years. Approximately 30% of these cases comes from the departments included in the study. Therefore, approximately 2,200 cases a year could be potentially eligible, around 6,600 during the accrual period (3 years). We enrolled 962 patients, around 15% of the estimated eligible population. We have not transferred this information to the flow chart because we do not have the exact number of potential eligible populations for the departments included nor the exact number of those who were invited or declined to participate.

Although only 20% of CCT-registered patients started to receive the transfers during the treatment period, the remaining 80% received it within the 12-month period following the completion. Interestingly, the benefit of the CCT was evident even with this important delay, reinforcing the potential effect of the program.

It is possible that a small number of patients was not registered into the program because they were deemed not eligible by the center health professional. However, the vast majority of patients not enrolled fulfilled the requirements but did not get registered due to different reasons, such as lack of awareness from health providers and patients about the CCT program or complex registration process and bureaucratic barriers. Since the information on the CCT was collected at the end of the study, we do not know the exact timing for when each patient was registered for the program. Nevertheless, all patients included in the program received the total amount corresponding to the full 6-month treatment period independently of the date of the registration.

The threat of multidrug-resistant TB (MDR-TB) continues to spread. It was recently estimated that in 2017, there were about 558,000 people developing TB resistant to rifampicin, of whom an estimated 458,000 had MDR-TB, defined as resistance to two first-line drugs, rifampicin and isoniazid, and 230,000 deaths globally. Argentina is one of the 5 countries in the Americas with a high number of estimated MDR-TB [40]. The main cause is poor adherence to TB treatment, and the consequences in public health are enormous: second-line drugs are less effective and treatments are much longer and substantially more expensive with more adverse effects, leading to a vicious circle of higher default rates and potentially catastrophic consequences [34]. Identifying interventions to improve compliance and reduce abandonment of treatment may represent a great contribution to reduce MDR-TB.

Sustainable Development Goal 1 (SDG 1) focuses on reducing poverty and expanding social protection. The link between poverty and TB has been well described, and evidence from ecological studies supports an association between increased social protection and decreased TB burden [35].

A recent study estimated the reduction in global TB incidence that could be obtained by reaching SDG 1’s targets of reducing poverty and expanding social protection. The model suggested that expanding social protection coverage may result in a reduction in TB incidence of 76% by 2035 [36]. These results align with the concept of Syndemics, an interesting approach to study, understand, and implement health research [37], refining the conventional frameworks that overlook the effects of social, political, and ecological factors and illuminating how macro-level social factors impact on health. As stated recently [38], purely biomedical or public health solutions are not enough to end the TB epidemic; countries must implement social policy strategies that can protect the patients and their contexts and prevent incomplete treatment since these interventions represent a critical and necessary investment.

We performed a formal evaluation of a health policy that aims to benefit not only individuals but also families and communities. We believe that our study provides valuable quantitative evidence of the effect of a CCT on sustained TB treatment adherence to project its effects and additional groundwork to support this strategy. The CCT appears to be a valuable health policy intervention to improve the control of TB in similar high-burden areas. The results of this study should encourage decision-makers to facilitate and promote a much wider implementation of these policies and increase the coverage to all TB patients and households living under vulnerable conditions.

Supporting information

Acknowledgments

We thank the HCFs, clinical and administrative staff, national TB programs, regional TB directors, and the patients for support of this study.

The authors’ views do not necessarily represent those of the NIAID.