A systematic review and meta-analysis of the catastrophic costs incurred by tuberculosis patients

screening Qualitative and quantitative


Method
We performed this systematic review and meta-analysis according to the Preferred Reporting Items of the Systematic Reviews and Meta-Analyses (PRISMA) guidelines 23 . Our research protocol was registered in PROSPERO (registration no. CRD42020221283).
Data source and search strategy. We searched EMBASE, Scopus, EBSCO, MEDLINE central/PubMed, ProQuest, SciELO, SAGE, Web of Science, and Google Scholar databases for articles without timeframe, geographical or language restrictions till November 20, 2020 by two authors (ShA & NZ). RMG and SA re-ran the data-base search to check the search strategy and number of citations reported. In addition, they checked the number of citations exported to the reference manager. Highly focused and sensitive search strategies were developed by RMG after approval of PubMed Help Disk. The search terms include "tuberculosis "OR "Mycobacterium tuberculosis" OR "Koch's disease" AND "catastrophic cost". (Supplementary Table 1). We searched reference lists from included publications by hand and contacted researchers who are expertise in these surveys to assist in identifying any relevant publications.

Study selection and eligibility criteria.
We included observational studies that reported the proportion of patients suffering from catastrophic costs during the intensive (first 2 or 8 months of treatment in drug sensitivity (DS) or MDR, respectively), or the continuation phases of TB treatment.
Four authors (AME, ShA, NZ, and EE) independently screened titles and abstracts for relevant studies. We excluded non-observational studies, case reports, editorial, reviews, letters, and studies that did not report income because the catastrophic costs could not be calculated or when catastrophic costs were not calculated at the individual level (when the total direct and indirect costs incurred by all patients divided by the total income).
Two authors (AME, HE) independently assessed the retrieved abstracts and the full texts of these studies to determine eligibility according to the inclusion criteria. Disagreements were resolved through discussion and consensus, or through consultation with a third reviewer (SA), who solved these differences based on study judgements.
Data extraction and analysis. Three authors (RMG, AME, HE) extracted the following data from eligible studies: country, study design, population criteria (age, sex, drug sensitive/resistant), treating facility (public/private sector) strategy of case finding (ACF/PCF), tool used to identify the catastrophic costs, and the catastrophic total costs and its determinants at different cut-off points.
The outcomes and definitions. The primary outcome was the proportion of patients with TB and their households who incurred catastrophic costs. It was defined as the total direct and indirect costs because of TB reaching or exceeding 20% of the patient's or household's pre-treatment annual income 6 . Additionally, we addressed the main predictors of catastrophic costs and different coping strategies. Finally, we assessed the catastrophic costs among patients according to their drug sensitivity as DS or MDR (with or without HIV) and strategy of case finding (ACF versus PCF). www.nature.com/scientificreports/ Secondary outcomes were the proportion of the direct costs to the total costs of TB treatment among DS or MDR, with or without HIV, catastrophic health expenditure (CHE; defined as the direct costs that reach or exceed 40% of patient's capacity to pay or 10% of their household income) 24 , and the different coping strategies.
Publication bias. We assessed publication bias by visual inspection of funnel plots and Egger's regression test.
Statistical analysis. Owing to the heterogeneity between studies, the proportion of catastrophic costs among patients with TB was pooled using the random effects model. Owing to the heterogeneity between studies, the proportion of catastrophic costs among patients with TB was pooled using the random effects model 26 .
Assessment of heterogeneity. Heterogeneity was assessed using the chi-square test on n-1 degrees of freedom, with an alpha of 0.05 considered for statistical significance and Cochrane-I-squared (I 2 ) statistic. I 2 values were classified as follows: 0 to 40%, might not be important; 30% to 60%, may represent moderate heterogeneity; 50 to 90%, may represent substantial heterogeneity; 75% to 100%, considerable heterogeneity 27 . Sources of heterogeneity, for identifying the possible effect modifiers on the pooled analyses, were explored using the following techniques: • Find-out outliers: If the study's confidence interval does not coincide with that of the pooled effect, it is considered an outlier. The size of the outlier has a substantial effect, and it deviates considerably from the overall effect. High-sampling-error studies vary significantly from the pooled result. However, because the confidence intervals of such studies are wide, there is a greater chance that the confidence intervals may overlap with one of the pooled effects. This basic outlier elimination technique is implemented using the find outliers function (dmetar) package. It seeks outlying studies in a (meta) item, eliminates them, and then recalculates the result (Supplementary Figure 1). • Sensitivity analysis: We used the metafor R tool to conduct a leave-one-out sensitivity analysis. In this method, we recalculate the meta-analysis results K times; each time excluding one study. The influence () function includes a set of leave-one-out diagnostic tests that help identify of influential studies. This analysis also includes a categorization of what is regarded as influential. We used I 2 to sort the studies in the plot. We identified studies with the highest heterogeneity and the final heterogeneity after excluding these studies (Supplementary Figure 2). We also created a Baujat plot, which compares the total heterogeneity contribution of each study to its effect on the aggregated outcome 26,28 (Supplementary Figure 3). • Graphic Display of Heterogeneity (GOSH) plots 29 : we fit the same meta-analysis model to all possible subsets of our included studies. In contrast to the leave-one-out method, we did not only fit K models, but also modelled for all 2 k − 1possible study combinations (Supplementary Figure 4).

Subgroup analysis.
We categorized the catastrophic costs at 20% for ACF and PCF patients, according to the country where the studies were conducted (inside/outside) India.

Meta-regression.
We studied the impact of the country where the survey was conducted (high versus low incidence of TB) 30 , quality of the study, sex, and population criteria (DS, drug resistant with or without HIV) on the size effect of studies to explain the substantial heterogeneity.

Secondary outcomes
Proportion of direct costs to the total costs. Direct to total costs among drug sensitive. The proportion of the mean direct costs to the mean total costs were addressed in six studies; the pooled proportion of direct to total costs at catastrophic costs of 20% were not calculated because of high heterogeneity. The proportion was variable; Tomeny et al. 17 and Timire et al. 52 reported that catastrophic costs were 41% and 43%, respectively. However, a higher proportion (52%) was reported by Chittamany et al. 37 and Nhung et al. 32 . Two other extreme values were reported by Fuady et al. 34 and Muttamba et al. 35 (33% and 65%, respectively).
Direct to total costs among multidrug resistance. The proportion of the mean direct costs to the mean total costs at cut-off point of 20% was addressed in seven studies ranged from 26% in Chittamany et al. 37 to 93% in Yang   32 . The pooled proportion of mean direct to total costs was difficult to assess because of the substantial heterogeneity which was not explained even after a meta-regression analysis.
Pooled proportion of direct costs to total costs in the case of ACF. The pooled proportion of the mean direct costs to the mean total costs was addressed in four studies and was (25%, 95% CI [16-37%]), I 2 = 83%. After conducting leave one out sensitivity analysis, the study of Gurung et al. 42 , was removed, the pooled proportion dropped to (29%, 95% C1 [20-41%]), I 2 = 55%. (Table 7).
Pooled proportion of direct costs to total costs in case of passive case finding (PCF). The pooled proportion of the mean direct costs to the mean total costs was addressed in four studies 19,20,42,50 and was (37%, 95% C1 [31-42%]), I 2 = 0% ( Table 8).

Proportion of direct costs to total costs in the case of HIV and TB co-infection.
The proportion of the direct costs to the total costs were addressed in two studies. Mudzengi et al. 18 showed that the proportion of mean direct costs to the mean total costs was 30% among HIV and TB co-infection patients, while a higher proportion (59%) was reported by Chittamany et al. 37 . We couldn't pool these studies because of the unexplained heterogeneity. In-between groups P < 0.0001  www.nature.com/scientificreports/ Proportion of mean direct costs to total costs. The pooled proportion of the mean direct costs to the mean total costs was addressed in 13 studies, which ranged from 4 to 87% (Supplementary Table 2).
CHE at 10% & capacity to pay at 40%. There were six studies calculated the CHE 10% and the capacity to pay (CTP) 40%.

Discussion
Our meta-analysis showed that the proportion of patients facing catastrophic costs at a cut-off point of 20% Catastrophic costs. The costs incurred by TB on some patients may be catastrophic and minimal for others. This is based on the household annual income. In the current study, we included 28 studies that addressed catastrophic costs among patients with TB at different thresholds points (30%, 25%, 20%, 10%, and 5%). Despite the absence of robust evidence on the sensitivity of the cut-off point at 20% to reflect the catastrophic costs,   www.nature.com/scientificreports/ regardless of whether patients are drug sensitive or resistant; Fuady et al. 14 established 15% and 30% as more consistent cut-off points for treatment adherence and success, respectively. In the current study, the proportion of TB-household patients facing catastrophic costs was 39%, which was considered very high compared with the targeted sustained developmental goals in 2020 (0%), thus more efforts and activities should be directed to reduce these costs. Diagnosis and treatment are provided free in many of the included countries under the umbrella pool of NTP; however, the treatment related expenditure is still very high. Yadav et al. 54 illustrated that even with free services for TB care, 21.3% of the patients were exposed to hardship financing, thus recommending more innovative ways to increase the supported coverage of TB treatment in the countries. The study also suggests the use of hardship financing as an index to measure the effectiveness of TB control program. It is crucial to decrease the burden of catastrophic costs among patients with TB, as it results in poorer treatment outcome. Patients suffering from catastrophic costs had 2-4 times higher odds of treatment failure than those who do not 14 . This could be explained by the reduced access to treating health facility and treatment completion. Regarding the coping costs, the majority of household's resort to different coping strategies to deal with the increased out-of-pocket costs and to compensate for the consequences of income loss. Those coping strategies include selling a property or livestock, taking loans, pledging jewels, dropping their children out of school, and cutting down their consumption to below basic needs 11 . Despite pooling of these studies' outcome yielded substantial heterogeneity, the current study found that 51.57% of heterogeneity was mainly because of two predictors; the first was that some studies estimated catastrophic costs of DS and patients with MDR with or without HIV together. This factor played a major role in the heterogeneity, as it was clear that the catastrophic costs were dramatically higher among patients with HIV. The second predictor was the classification of the country where the study was conducted 30 . Two-thirds of the new cases of TB are reported in eight countries of the world, with India being the highest, followed by Indonesia, China, the Philippines, Pakistan, Nigeria, Bangladesh and South Africa. Consequently, we sub-grouped the studies according to the country where they were conducted; countries with high versus low TB incidence. In meta-regression analysis, the country where the study was conducted was the second major determinant of the different size effect. The reported high incidence of catastrophic costs in many countries raised the need for social protection interventions. The most common social protection intervention is the cash transfer or cash assistance, which has already been implemented in many countries across the world, either conditionally or unconditionally 55 . Thus, it is supposed that the household can get better access to treatment and food. Other social protection interventions include disability grants, food baskets (food assistance), food or travel vouchers and social insurance 11 . Many countries have implemented reimbursement programs to help patients with TB to cope with the disease costs. However, these programs prioritize poorer and MDR 56 . The effect of this intervention is questionable. At a cut-off point of 20%, two studies have applied and calculated a catastrophic costs before and after reimbursement. Lue et al. 45 reported a minimum change in the proportion of catastrophic costs; before reimbursement, the catastrophic costs were (22%) and declined to 19% after the reimbursement. In contrast, Fuady et al. 57 showed a higher change in the proportion of catastrophic costs after the reimbursement. The intervention program effectively decreased catastrophic costs from 44 to 13%. Regarding cash transfer, Wingfield et al. 53 reported that the proportion of TB households suffering from catastrophic costs was 30% and 42% among intervention and control groups, respectively. These findings indicate that this social support is not enough to mitigate the impact of TB. Consequently, households of TB patients should receive sufficient financial support that covers indirect costs (job lost) and direct costs (transportation, food, accommodation) 58 . Such social support should be proportionate to the income lost because of the high variability of the pre-treatment income. We speculate that developing newer treatment guidelines for TB of a shorter duration would be beneficial. At the bottom, providing free medication is insufficient to prevent the catastrophic costs. TB patients should receive transport vouchers, reimbursement schemes, and food assistance to reduce or compensate for such catastrophic costs. Furthermore, decentralization of patient supervision (including directly observed therapy), for example, through community-based or workplace-based treatment 59 , can reduce transport costs and income loss for patients 60 .
As expected, the catastrophic costs among MDR were higher than among DS, as DS patients receive treatment for shorter duration (6 months only), while MDR treatment extends to 24 months. Additional cost is incurred by MDR patients, such as the cost related to prolonged work absenteeism, need for daily injection, exposure to more side effects, and need for more investigation 61 .
Direct costs to total costs. The mean total direct costs to the mean total costs were lower than the mean indirect costs among drug sensitive patients, HIV co-infected patients, while it was higher among drug resistant patients. This finding is essential to be considered when reimbursement strategies are implemented. Stakeholders should know which part of patient costs should be compensated. The direct costs dropped significantly if the ACF strategy was adopted instead of the PCF (29% to 37%) respectively.

Determinant of catastrophic costs.
Recognizing the determinants of catastrophic costs could provide an insight into approaches for mitigating catastrophic costs among the vulnerable TB patients and their households. The epidemiological consequence of TB is directly related to a country's socioeconomic profile. TB vulnerability is determined by biological variables (e.g., malnutrition, HIV infection, and age) and social factors (e.g., poor housing conditions, high population density, inhumane working conditions, and a lack of access to health services). Under many circumstances, numerous vulnerabilities occur simultaneously 62 . In this study, the main determinants of catastrophic costs were income loss as an impact of being diseased and food and nutritional supplements other than the patients' regular diet habit [37][38][39] . In addition, travel and transportation affected the direct non-medical costs, thus increased the suffering of patients with TB 35 . Age also affected the proportion of patients with TB suffering from catastrophic costs, whether young 47 or old 36  www.nature.com/scientificreports/ that delayed treatment initiation was a major predictor of catastrophic costs. Approximately 24% of individuals with catastrophic expenses waited for more than four weeks following the onset of symptoms to start treatment. Severe symptoms, prolonged hospitalization, more expensive non-TB medication, or even more frequent visits to the facilities may explain why delayed treatment initiation was related. Health and social protection investments have minimized the negative health effects of TB. Moreira et al. 62 emphasized the relevance of public social protection programs in mitigating the consequences of TB indicators in the pursuit of TB elimination.
Catastrophic health expenditure. Out of the 28 studies, only six studies have been included with a clear measurement of the CHE (at cut-off point of 10% of their income and at cut-off point of 40% of their CTP). It was clear that many studies ignored CHE, despite its importance to understand the impact of these costs on treatment outcomes 45 . Two studies assessed the effect of reimbursements intervention on the CHE. Xiang et al. 64 reported a 8% reduction in CHE, however, this reduction was not statistically significant. Similarly, Zhou et al. 65 reported that the effect of reimbursement on CHE was minimal; only 12% reduction in CHE was achieved. To decrease the catastrophic expenditures national health financing systems must be designed and implemented, not only to allow people to access services when they are needed but also to protect households from financial catastrophe, by reducing out-of-pocket spending. Eventually, prepayment mechanisms should be developed, for instance, social health insurance, tax-based financing of health care, or some mix of prepayment mechanisms such as efficient reimbursement or cash intervention 66 . Strengths and limitations of the study. Our study has several strengths and limitations. The strengths include the wide sensitive search strategy and multiple studied outcomes. The limitations of this study was that different cut-off points were established by different studies to estimate the proportion of the households facing catastrophic costs using different tools. Second, a major challenge was that different studies estimated catastrophic costs due to TB, regardless of drug sensitivity (DS, MDR), co-infection with HIV, case finding strategy (ACF, and PCF). Third, all studies included subjects with confirmed TB. Costs for those ill patients with undiagnosed TB may add much to the already estimated values. Fourth, many of the included studies used the WHO cost survey tool which included patients only treated in the NTP, omitting patients treated in private sectors who represent a considerable proportion. Fifth, owing to the observational nature of the studies included, there was a risk of recall bias as well as a chance of reporting inaccurate significant relationships due to confounders. Sixth, the degree of heterogeneity across the studies was likewise substantial; so, the random effects method was used to generate the pooled data. In addition, we adopted several techniques to overcome it like meta-regression and subgroup analysis. Finally, the quality of a meta-analysis is determined by the quality of the included research; here, the quality of the majority of the included studies was graded as satisfactory or good.

Conclusion
Regarding future global policy, our study provides an evidence for the high proportion of TB patients who are still facing catastrophic costs despite the free TB treatment policy. The proportion of patients facing catastrophic costs varies according to the type of TB, which is the lowest among DS, higher in MDR, and the highest among those with concomitant infection with HIV. Patients exposed to ACF incurred lower costs than those exposed to PCF. The direct costs (medical &non-medical) related to TB is not the only major contributor to the catastrophic costs, but indirect costs (Job and productivity lost) also represent a major contributor that should not be ignored. Overall, this study paves the way to effective cost mitigation in the context of the End TB Strategy. Effective management of the predictors of catastophic costs will eventually contribute to better community, clinical, and financial outcomes. It is clear that the global health system must make more efforts to achieve the zero catastrophic costs for TB by 2030. Future research should assess the effectiveness of reimbursement for TB patients on the reduction of the proportion of patients who face catastrophic costs. Furthermore, in an attempt to reduce the costs incurred by TB patients, researchers should develop more reliable diagnostic tools to reduce patients' need for frequent visits to healthcare facilities. They also have to study the impact of educational programs on TB patients' compliance with the prescribed medicine to lower the retreatment rates. Finally, NTP should monitor the financial and social status of patients treated and intervene as early as possible to protect them from incurring these catastrophic costs.

Data availability
Data are available upon request by contacting the first author.