Catastrophic costs among tuberculosis patients in Zimbabwe: a national health facility-based survey

Background: Tuberculosis (TB) has biological and socio-economic dimensions. The socio-economic impact of TB may plunge households into nancial catastrophes. The End TB strategy seeks to mitigate socio-economic factors which may act as barriers to accessing TB services. Countries have been encouraged to provide baseline estimates of catastrophic costs by 2020. We sought to determine the prevalence, risk factors and major drivers of catastrophic costs among TB patients in Zimbabwe. Methods: We conducted a nationally representative, health facility-based survey with random cluster sampling among TB patients. We enrolled patients with drug susceptible (DS-TB) and drug resistant TB (DR-TB) consecutively. We administered a standardised questionnaire to capture the costs incurred as well as lost income due to TB illness. Catastrophic costs were measured at a threshold of 20%. We did a sensitivity analysis of indirect costs using the time lost by patients in each phase of treatment. We used multivariable logistic regression to determine the risk factors for experiencing catastrophic costs. Results: A total of 841 patients were enrolled in the survey, weighted to 900 during data analysis. There were 500 (56%) males and 46 (6%) DR-TB patients. Thirty-ve (72%) DR-TB patients were HIV co-infected. Overall, 80% (95% CI:77-82%) of TB patients and their households experienced catastrophic costs. The major cost drivers pre-TB diagnosis were direct medical costs. Nutritional supplements were the major cost driver post-TB diagnosis, (median cost US$360 (IQR: 240-600). Post-TB diagnosis costs were three-times higher among DR-TB [US$1,659 (653-2,787)] versus drug susceptible TB (DS-TB) affected households [US$537 (204-1,134)]. Income loss was ve-times higher among DR-TB versus DS-TB patients. In multivariable analysis, household wealth was the only covariate that remained signicantly associated with catastrophic costs: the poorest households had sixteen times the odds of incurring catastrophic costs compared to wealthiest households [adjusted odds ratio (aOR:15.7 95% CI:7.5-33.1)]. Conclusion: The majority of TB patients enrolled in care experienced catastrophic costs, with the highest

Background Tuberculosis (TB) is an archetypical poverty-related disease, with poverty both a cause and a consequence of TB. In any country (low, middle or high-income), TB affects the poorest and most marginalised groups of society. (1,2) In 2018 approximately 10 million people developed TB disease globally and 1.6 million of them died, making TB one of the top ten killer diseases worldwide and the leading cause from a single infectious agent. (1) In sub Saharan Africa, TB has a powerful synergy with the human immune-de ciency virus (HIV) aggravating the clinical and economic impact of disease at the patient and household level. The economic impact of TB acts as a barrier to both access and adherence to TB treatment services.
Traditional biomedical approaches to TB control have not yielded substantial reductions in the global incidence of TB, and it was hypothesised that TB elimination would not be achievable until after the year 2182. (7) Biomedical approaches tend to con ne the responsibility for TB treatment within the healthcare sector, ignoring the social determinants of TB disease, which lie outside the healthcare sector. The End TB Strategy has encouraged thinking beyond the biomedical model by including one target speci cally relating to costs (no TB patients or households should experience catastrophic costs by 2020). (7) Total TB-related costs are de ned as catastrophic if they exceed 20% of a household's annual income.
Catastrophic costs are a public health challenge requiring urgent attention. (4) Such costs may plunge households into poverty and nancial catastrophes. (5) For this reason, countries are encouraged to set baseline measurements of catastrophic costs by 2020.The measurements are based on three types of costs: direct medical, direct non-medical and indirect costs such as income loss. Direct medical costs include money spent on consultations, laboratory tests, and hospitalisation. Direct non-medical costs are money spent on transport and food during health seeking. Income loss is money foregone by the patient or carers during illness.
Global efforts to ameliorate catastrophic costs lie in i) provision of free TB treatment ii) decentralisation of TB services to ensure equity of access and iii) advocacy for social protection and universal health coverage by the United Nations, national and international stakeholders. Despite interventions aimed at cushioning TB patients against direct medical costs, surveys in Africa and Asia have shown high prevalence of catastrophic costs especially among i) drug resistant TB (DR-TB) patients, ii) poorest households, iii) cases where the patients were breadwinners and iv) those co-infected with HIV. (8, 9,[11][12][13] In 2016, Zimbabwe embraced the End TB targets of eliminating catastrophic costs by 2020. The national TB control programme (NTP) has decentralised TB services, provided cash transfers to DR-TB patients and adopted active case nding to detect TB cases early. However, the country had no baseline measure of catastrophic costs or the major drivers of such costs. We therefore sought to determine the prevalence, risk factors and major drivers of catastrophic costs among patients on TB treatment in Zimbabwe.

Study design
A nationally representative, health facility-based survey with random cluster sampling among TB patients across Zimbabwe.

Setting
Zimbabwe is a southern African country with an estimated population of 14.7 million. (https://www.worldometers.info/world-population/zimbabwe-population/). Classi ed as a low income country, it has suffered from an economic and humanitarian crisis for much of the last decade.
Zimbabwe belongs to the 14 countries with a triple-burden of TB, TB/HIV and multi-drug resistant TB.

Study population
Patients of all age groups who were on treatment for drug susceptible TB (DS-TB) or DR-TB, and who attended their scheduled appointments within the sampled health facilities from 23 July to 31 August 2018 were eligible for inclusion in the study.

Study procedures
Sample size calculation, sampling and patient enrolment The sample was based on 26,677 TB patients noti ed in Zimbabwe in 2016. We assumed an absolute precision of 5% and a priori guess of 50% for the prevalence of households experiencing catastrophic costs. We used the standard formula for sample size calculation for a cluster sampled TB prevalence survey. (18) After factoring a design effect of 2.0, the sample size was 780 patients across 60 clusters, each contributing 13 patients. After factoring in an estimated non-response rate of 10%, the sample was adjusted to 900 patients.
Cluster sampling was used to select health facilities (clusters). First, a list of clusters and their corresponding 2016 noti cations was compiled. The number of TB noti cations per cluster was used as a proxy for the size of clusters. Clusters that noti ed < 10 patients were merged with adjacent clusters.
Second, cumulative noti cations were compiled and 60 clusters were selected by a probability proportional to size sampling method using a randomly de ned starting point and sampling interval.
Patients were recruited consecutively when they attended their scheduled appointments. Interviews were postponed for patients who had not been on treatment for two weeks in their current treatment phase (intensive or continuation phase). All clinical and economic data were collected for the respective phase only.
The study questionnaire was adapted from the WHO generic instrument and created in CSPro® (Census Bureau, USA). Data collectors, one per facility, were trained by the NTP and partner organisations. They comprised TB focal nurses and Environmental Health Technicians. At the end of each day data from tablets were synchronised electronically with a central server at central level. Zimbabwe National Statistics Agency staff monitored the server and promptly highlighted errors for clari cation. Periodic data quality checks and support visits were conducted by the steering committee. A WhatsApp group for data collectors and steering committee members was created to aid in addressing operational challenges, mostly related to patient recruitment and syncing electronic records in real-time.

Data Variables, Source Of Data And Data Collection
Socio-demographic and clinical data (age, HIV status, type of TB patient) were extracted from treatment registers and patient treatment booklets prior to the interview. Data on hospital visits, costs incurred, household assets and coping strategies were collected by trained data collectors during face-to-face interviews with patients. All the interviews were conducted in separate rooms within health facilities to ensure con dentiality.

Data analysis
Anonymised data were exported to Stata version 15.0 (StataCorp, College Station, TX, USA) for cleaning and analysis. Categorical variables were summarised using frequencies (proportions), while continuous variables were summarised using means [standard deviations (SD)] or medians [inter-quartile ranges (IQR)] as appropriate strati ed by DR status. We summed up the direct medical costs (consultations fees, laboratory tests) and direct non-medical costs (transport, food). Costs that were in South African Rand were converted to USD using the prevailing conversion rate obtained from the Oanda currency converter (http://www.oanda.com). Productivity losses due to TB treatment were estimated using the output approach, where the difference in monthly income before and after TB diagnosis was extrapolated over the treatment period. A sensitivity analysis of indirect costs estimation was done using valuation of the time lost by the patient in each phase of treatment. To estimate patient costs for the entire TB episode, including costs for all phases of treatment, we extrapolated costs based on data from patients in other phases of illness. We used the approach recommended by WHO, whereby we replaced missing cost data with median costs of the phase of illness among those in that phase with available data. Comparisons between categorical variables were done using the chi-square test. Multivariable logistic regression was used to determine risk factors for patients experiencing catastrophic costs after adjusting for sex, age, DR status; treatment phase, HIV status, breadwinner status, household income and location of health facility.
The level of signi cance was set at P < 0.05.

Results
A total of 860 patients (96% of the target sample size) were reached and consented (Fig. 1). Of these, 19 records were excluded from analysis for being on treatment for < 14 days (17) and for failure to complete the interview (two). Overall, the response rate was 841 (93%). The gure was 900 after factoring in nonresponse weights.
Of the 900 patients, 851 (94%) had DS-TB (Table 1). The mean age was 36.9 years and 500 (56%) were males. A greater proportion of DR-TB than DS-TB patients were in the continuation phase (66% vs 56%) and were HIV-positive (72% vs 61%). Almost all patients were new, rather than retreatment. DR-TB patients were more likely to live in urban locations than DS-TB patients (69% vs 59%). The proportion of patients/households who experienced catastrophic costs was 80% (95% CI: 77-82) (Fig. 2). The prevalence of catastrophic costs was 90% among DR-TB patients and 79% among DS-TB patients, p = 0.06 (Table 2). Overall, 95% of patients in the poorest income quintile experienced catastrophic costs. Income quintile was strongly associated with catastrophic costs in a dose-response relationship after adjusting for other variables. Compared to the wealthiest group, the poorest households had higher odds of incurring catastrophic costs, [adjusted odds ratio (aOR): 15.7 (7.5-33.1)]. Sex, age, type of TB, treatment phase, treatment delay (≥ four weeks), HIV status, being a breadwinner and location of health facility were not associated with catastrophic costs in either univariable or multivariable analysis.  Overall, the total median cost per TB episode was US$1,247 (IQR:545-2405) ( Table 3 ). The total costs incurred by DR-TB patients were three times higher than those of DS-TB patients. Costs as a proportion of total costs were: non-medical costs (51%); indirect costs (36%) and medical costs (13%

Discussion
We found a high prevalence of catastrophic costs among TB patients/households. The poorest households experienced the highest catastrophic costs. There were no differences in proportions of DR-TB and DS-TB households which incurred catastrophic costs. The major drivers of catastrophic costs were direct non-medical and indirect costs related to productivity loss. Indirect costs were ve-times higher among DR-TB compared to DS-TB patients.
Previous surveys done in Africa and Asia have shown that TB patients incur huge catastrophic costs despite free TB treatment. (8,9,11,12) Our study provided additional evidence to this nding. However, catastrophic costs were not homogeneous among the different groups as the poorest households were disproportionally affected and risk of catastrophic costs increased as wealth quintile decreased (5,8,11,12) This makes intuitive sense, and may be attributed to reduced resilience to external shocks such as TB. Unlike studies done elsewhere, we did not nd a difference in catastrophic costs by DR status. We had low numbers of DR-TB patients, and our study may not have been su ciently powered to detect the difference. Also, in our context even DS-TB patients experienced higher catastrophic costs than overall costs reported from other low-and middle income countries. (8, 9,12,13) High catastrophic costs may negatively impact on both access and adherence to TB treatment.
The major drivers of catastrophic costs lay outside the healthcare sector, a consistent nding with studies wherein non-medical costs were reported to account for up to 80% of catastrophic costs. (7,12,18) Our study highlights an urgent need to address socio-economic cost drivers such as income loss due to loss of productivity time, travel costs and nutritional supplements. These social determinants of TB have a major impact on TB health outcomes. Future studies should unravel both the type and source of nutritional supplements that are purchased by TB patients in Zimbabwe.
Patients on DR-TB treatment experienced far higher income losses than DS-TB patients in this study. The reasons could be three-fold: lengthy treatment requiring frequent visits to health facilities; loss of productivity time since TB affects mostly the economically productive age groups (25-44 years); lack of income replacement due to the informal nature of businesses in Zimbabwe and lack of social protection.
Even in contexts where social protection is available, income loss poses the greatest nancial risk to TB patients. (5)  costs. This was followed with a social protection mapping exercise which identi ed barriers to accessing social protection such as lack of knowledge about availability of services and cumbersome registration processes. (21) This study is strengthened by the fact that we recruited patients consecutively to minimise selection bias.
We minimised data entry errors through use of validated electronic questionnaires with check functions.
However, there were limitations in that patients who sought TB care outside Zimbabwe were not represented in this survey. Moreover, we interviewed the patients once and had to estimate most of the costs. Recall bias could affect cost estimates for the pre-treatment period, leading to under-or overestimation of the costs. However, patients may not forget about the painful experiences they went through especially those related to selling productive assets. We minimised recall bias by interviewing persons in the intensive phase about diagnostic costs and the costs that were incurred prior to diagnosis. We could not capture both direct and indirect costs after treatment outcomes (including burial costs).
These costs can extend well beyond the treatment period, even for people who are declared cured from TB.
Conclusion TB patients and their households incur huge catastrophic costs in Zimbabwe despite free TB treatment.
The major cost drivers could be ameliorated through social protection and universal health coverage. A multi-sectoral approach to TB control holds great promise to reducing catastrophic costs due to TB in