Tom WingfieldpersName, Delia BocciapersName, Marco TovarpersName, Arquímedes GavinopersName, Karine ZevallospersName, Rosario MontoyapersName, Knut LönnrothpersName, Carlton A. EvanspersName (2014date). “Defining catastrophic costs and comparing their importance for adverse tuberculosis outcome with multi-drug resistance: a prospective cohort study, Peru..” PLoS Medicinetitle·j.
Object of study — tuberculosis-related household costs and their association with adverse tuberculosisterm·key outcome
Research question — whether catastrophic coststerm·dom of TB-affected households are independently associated with adverse TB outcometerm
Contribution — the first evidence-based threshold for clinically relevant catastrophic coststerm·dom, with a method to test it elsewhere
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | hidden direct and indirect costs borne by households despite free tuberculosis treatment — “Patients and their households may incur hidden costs, be they direct “out of pocket” expenses such as for transport, symptom-relieving medicines, or additional food, or indirect expenses associated with lost income [8] – [12] .” | |
| claim-type | existence — “Having MDR TB and incurring catastrophic costs were independently associated with adverse TB outcome, with a similar adjusted population attributable fraction for adverse TB outcome.” | |
| method — evidence | — “We conducted a prospective cohort study of TB patients and a baseline case-control study comparing them with healthy controls.” | |
| object of study — evidence | — “The aims of the study were to better characterise TB-related costs and their association with adverse TB outcome, and to contribute to an evidence-based definition of catastrophic costs that is both clinically and financially relevant.” | |
| research question — evidence | — “The study hypothesis was that catastrophic costs of TB-affected households are independently associated with adverse TB outcome in TB patients.” | |
| contribution — evidence | — “Thus, our findings provide, to our knowledge, the first evidence-based threshold for clinically relevant catastrophic costs, and demonstrate a methodology to assess the generalizability of this threshold in other settings.” | |
| mechanism | catastrophic costs Catastrophic costs in poor households can cause financial shock: cutting consumption below minimum needs, selling assets and taking children out of education. — “Catastrophic costs in poor households can lead to financial shock: families reducing consumption below minimum needs, selling assets, and taking children out of education.” | mechanisms[0] |
| sense | catastrophic costs (earlier operational thresholds) - total costs of 10%-25% of annual household income, or 40% or more of capacity to pay — “Others have used operational thresholds of total costs of 10%–25% of a household's annual income [16] , [19] , [20] or 40% or more of a household's “capacity to pay” [21] , [22] .” | senses[0] |
| sense | medical poverty trap - as incomes fall, costs as a share of income rise — “In agreement with other findings, our results indicate that costs as a proportion of the same household's income indicate economic challenge better than actual monetary expenditures: this finding highlights the “medical poverty trap”, that as incomes decrease, proportional costs increase [6] , [45] .” | senses[1] |
| relation | Lost income —same-as→ indirect expenses — “Lost income” (indirect expenses) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles. — ““Lost income” (indirect expenses) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles.” | concept_edges[0] |
| weight | Lost income —same-as→ indirect expenses — 0.85 — ““Lost income” (indirect expenses) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles.” | concept_edges[0] |
| relation | multi-drug-resistant —same-as→ MDR — We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistant (MDR) and non-MDR TB in impoverished shantytowns surrounding Lima, Peru. — “We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistant (MDR) and non-MDR TB in impoverished shantytowns surrounding Lima, Peru.” | concept_edges[1] |
| weight | multi-drug-resistant —same-as→ MDR — 0.9 — “We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistant (MDR) and non-MDR TB in impoverished shantytowns surrounding Lima, Peru.” | concept_edges[1] |
| effect direction | Poverty —causes→ TB risk — increases — “Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] .” | concept_edges[2] |
| relation | Poverty —causes→ TB risk — Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] . — “Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] .” | concept_edges[2] |
| weight | Poverty —causes→ TB risk — 0.75 — “Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] .” | concept_edges[2] |
| replication | original | |
| limitation | The financial effects of MDR TB may be underestimated because cost questionnaires ran for only 6 months while MDR treatment lasts 18 months or more. — “Third, we may have underestimated the financial effects of MDR TB because our questionnaires quantifying costs continued for only 6 mo, whereas patients with MDR TB are usually treated for 18 mo or more.” | limitations[0] |
| limitation | The methodology needs repeating in other settings to test external validity. — “Finally, our research demonstrates a new methodology that should be repeated in other settings to assess the external validity of our findings.” | limitations[1] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | The unadjusted population attributable fraction of adverse TB outcomes explained by catastrophic costs was 26%, similar to that of MDR TB (23%). — “The unadjusted population attributable fraction of adverse TB outcomes explained by catastrophic costs was 26% (95% CI = 14%–36%), similar to that of MDR TB (23% [95% CI = 17%–28%]).” | content_tags.findings[0] |
| quantity | 345 — — “Catastrophic costs were incurred by 345 households (39%).” | content_tags.quantities[0] |
| key-terms | catastrophic costs, tuberculosis, multi-drug resistance, lost income, direct expenses, poverty, social protection, adverse TB outcome, population attributable fraction, Peru |
| concept | catastrophic costs — total costs greater than or equal to 20% of annual income for TB-affected households — “Our novel findings also define an evidence-based threshold for catastrophic costs of total costs greater than or equal to 20% of annual income for TB-affected households.” | concepts[0] |
| concept | adverse TB outcome — death during treatment, treatment abandonment, treatment failure, or recurrence — “We defined adverse TB outcome as death during treatment, treatment abandonment, treatment failure, or recurrence.” | concepts[1] |
| concept | good TB outcome — cure without recurrence — “We defined good TB outcome as cure without recurrence.” | concepts[2] |
| concept | Lost income — the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview — ““Lost income” (indirect expenses) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles.” | concepts[3] |
| concept | indirect expenses — not defined in this paper's own words; the other name the paper gives to income lost through illness — ““Lost income” (indirect expenses) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles.” | concepts[4] |
| concept | Total expenses — direct expenses plus lost income — ““Total expenses” were direct expenses plus lost income.” | concepts[5] |
| concept | population attributable fraction — the proportion of adverse TB outcomes that would be averted by eliminating that exposure — “The population attributable fraction of an exposure was interpreted as the proportion of adverse TB outcomes that would be averted by eliminating that exposure, both unadjusted and adjusted for known confounding factors.” | concepts[6] |
| concept | multi-drug-resistant — not defined in this paper's own words; tuberculosis that several first-line drugs no longer kill — “We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistant (MDR) and non-MDR TB in impoverished shantytowns surrounding Lima, Peru.” | concepts[7] |
| concept | MDR — multi-drug-resistant — “We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistant (MDR) and non-MDR TB in impoverished shantytowns surrounding Lima, Peru.” | concepts[8] |
| concept | Poverty — not defined in this paper's own words; a household's lack of money and assets — “Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] .” | concepts[9] |
| concept | TB risk — not defined in this paper's own words; the chance of developing tuberculosis — “Poverty increases TB risk [5] , and TB exacerbates poverty, affecting the most economically productive age group [6] – [8] .” | concepts[10] |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
Defining catastrophic coststerm·dom and comparing their importance for adverse tuberculosisterm·key outcome with multi-drug resistanceterm·key: a prospective cohort study, Peruterm·key.
Abstract Tom WingfieldpersName and colleagues investigate the relationship between catastrophic coststerm·dom and tuberculosisterm·key outcomes for patients receiving free tuberculosisterm·key care in Peruterm·key.
Please see later in the article for the Editors' Summary Introduction Tuberculosisterm·key (TB) disease kills 1.4 million per year and remains a major global health problem [1] . Many low- and middle-income countries are unlikely to meet the Millennium Development Goals for reduction of TB disease prevalence and mortality [1] . This is due in part to poorer people experiencing inequitable healthcare provision and access [2] and suffering a disproportionate burden of morbidity and mortality from TB disease [3] , [4] .
Povertyterm increases TB riskterm [5] , and TB exacerbates povertyterm, affecting the most economically productive age group [6] – [8] . Whilst many countries aim to offer “free” TB treatment to their patients, this free treatment may cover only some diagnostic tests and anti-mycobacterial medications. Patients and their households may incur hidden costs, be they direct “out of pocket” expenses such as for transport, symptom-relieving medicines, or additional food, or indirect expensesterm associated with lost incometerm [8] – [12] .
In its post-2015 Global Strategy and Targets for Tuberculosisterm·key Prevention, Care, and Control at the 67th World Health Assembly in May 2014, the World Health Organization adopted a target of eradicating catastrophic coststerm·dom for TB-affected families by 2035 [13] . However, hidden TB-related costs remain understudied, and consensus about defining catastrophic coststerm·dom is awaited [5] , [13] – [16] . Some catastrophic coststerm·dom definitions have incorporated symptoms of financial shock and coping mechanisms [17] , [18] . Others have used operational thresholds of total costs of 10%–25% of a household's annual income [16] , [19] , [20] or 40% or more of a household's “capacity to pay” [21] , [22] . Recently, concerns have been raised that the current approach of measuring catastrophic coststerm·dom using out-of-pocket payments is too narrow because it overlooks lost incometerm and consequently risks misinforming policy-makers [23] , [24] . Thus, there is an urgent need to improve indicators of financial risk to better inform health policy guidance [23] – [25] . However, although there is broad agreement that some vulnerable TB-affected households will require social protectionterm·key (such as socioeconomic support) to avoid catastrophic coststerm·dom, more evidence is needed to define such costs and characterise their importance [9] , [14] , [21] – [28] .
We prospectively quantified changes in income and hidden costs prior to and throughout treatment of patients with multi-drug-resistantterm·dom (MDRterm·dom) and non-MDR TB in impoverished shantytowns surrounding Lima, Peruterm·key. The aims of the study were to better characterise TB-related costs and their association with adverse TB outcometerm, and to contribute to an evidence-based definition of catastrophic coststerm·dom that is both clinically and financially relevant. The study hypothesis was that catastrophic coststerm·dom of TB-affected households are independently associated with adverse TB outcometerm in TB patients.
Methods Ethical Approval The internationally accredited ethical committee of the Universidad Peruana Cayetano Heredia approved the project. All interviewed participants gave written informed consent.
Study Design and Participants We conducted a prospective cohort study of TB patients and a baseline case-control study comparing them with healthy controls. From 26 October 2002 to 30 November 2009, in collaboration with the Peruvian National Tuberculosisterm·key Control Program, all consecutive patients with laboratory-proven pulmonary TB were invited to participate in the study. All interactions between the research team and the participants occurred during household visits. Until 30 November 2012, patients were followed-up for recurrent TB by monitoring TB treatment records and revisiting each household approximately every 3 y to enquire about TB diagnoses. From 21 December 2006 to 9 December 2007, control households were selected from an up-to-date satellite map using random number tables and were invited to participate during a household visit. In the case that all potential control participants in the household were unavailable or declined then the nearest neighbouring household was instead invited to participate. Controls were not matched to cases because the study aimed to characterise the effect of relevant exposures including sex, age, and socioeconomic position on the outcome variables of catastrophic coststerm·dom and adverse TB outcometerm. The inclusion criterion in both cases and controls was age more than 15 y. Exclusion criteria included declining or being unable to give informed written consent. Both for the cohort and the baseline case-control study, the sample size was opportunistic, and consequently no power calculations were performed.
Study Setting The study was conducted in Ventanilla, 16 peri-urban contiguous shantytowns in north Lima, Peruterm·key, with an estimated population of 277,895 people and frequent povertyterm (32% of inhabitants live on ≤US$1 per day). During the study period, the annual TB notification rate in Ventanilla was 162 new cases per 100,000 people per year, higher than the rest of the country, at 106 per 100,000 people annually [29] .
TB was treated by the National Tuberculosisterm·key Control Program in community health posts where sputum smear was offered free of charge to all patients, and chest radiographs to selected patients. TB patients received their anti-TB directly observed therapy (DOT) free of charge at their local health post, administered by the national TB program.
Variables Operational definitions of the key study variables (TB disease, TB treatment phases, TB adverse outcome, and TB costs) are summarised in Box 1 .
Box 1. Glossary of Operational Definitions of TB Disease, Treatment, Outcome, and Costs TB Disease MDRterm·dom TB: patients initially prescribed an MDRterm·dom treatment regimen or who had a sputum test positive for MDRterm·dom TB by the microscopic-observation drug-susceptibility (MODS) assay or the proportions assay Non-MDR TB: all patients recruited to the study not meeting the definition for MDRterm·dom TB TB Treatment Phases* Pre-treatment: the period of time from self-reported onset of TB-related symptoms until treatment initiation Intensive treatment phase: the first two consecutive months of TB treatment Continuation treatment phase: the four consecutive months immediately following the intensive treatment phase During treatment: the period of time spanning from the beginning of the intensive treatment phase to the end of continuation treatment phases Entire illness: the period of time from the onset of TB-related symptoms to the end of the continuation treatment phase TB Treatment Outcome Adverse TB outcometerm: patients who died during treatment (irrespective of cause), abandoned treatment, had treatment failure, or had recurrent TB disease within 30 mo of starting TB treatment Good TB outcometerm: patients who were declared cured by the TB program and had no recurrence of TB disease within 30 mo of starting treatment Undefined TB outcome: patients who were transferred by the national TB program to another health post outside of the study site or were lost to follow-up TB Costs Direct (“out of pocket”) expenses: the sum of the direct medical expenses and direct non-medical expenses Direct medical expenses: costs of medical examinations and medicines Direct non-medical expenses: costs of natural remedies, TB-care-related transport, extra food, and other miscellaneous expenses Lost incometerm (indirect expensesterm): the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) (a) from symptom onset until the recruitment interview and (b) from the previous interview date until subsequent interviews Total expensesterm: direct expensesterm·key plus lost incometerm Earnings: the monthly money actually received by the household Income: the monthly money that would have been earned by the household if it were not TB-affected (earnings plus lost incometerm) Catastrophic coststerm·dom threshold: the threshold at which total household expenses as a proportion of annual income were most strongly associated with adverse TB outcometerm; the strength of the association was assessed by the highest sensitivity, specificity, and population attributable fractionterm for adverse outcome *These treatment definitions apply to all TB patients, irrespective of whether they had MDRterm·dom or non-MDR TB.
Patients were defined as having MDRterm·dom TB if they were initially prescribed an MDRterm·dom TB treatment regimen or sputum testing was positive for MDRterm·dom TB by the microscopic-observation drug-susceptibility (MODS) assay or the proportions assay. All other patients recruited to the study were defined as having non-MDR TB.
For both patients with MDRterm·dom and patients with non-MDR TB, stages of treatment were operationally defined as follows: “pre-treatment” was from self-reported onset of TB-related symptoms until treatment initiation; “intensive treatment phase” was the first 2 mo of TB treatment; “continuation treatment phase” was the 4 mo immediately following the “intensive treatment phase”; “during treatment” was the period of time from the start of the “intensive treatment phase” to the end of the “continuation treatment phase”; and “entire illness” was the period of time from TB-related symptom onset to the end of the “continuation treatment phase”.
Early TB treatment outcome for each patient was assessed by the national TB program at the time of treatment cessation and was not influenced by this research. These early TB treatment outcome assessments were based on sputum microscopy results that are insensitive to treatment failure [30] – [32] . Therefore, we also collaborated with the national TB program in continuous surveillance of national TB program treatment records and revisited each patient in their home to check for TB recurrence, which we defined as TB retreatment within 30 mo from the date that treatment started (in most cases 2 y from treatment cessation). We defined good TB outcometerm as cure without recurrence. We defined adverse TB outcometerm as death during treatment, treatment abandonment, treatment failure, or recurrence. Patients who were transferred by the national TB program to another health post outside of the study site or were lost to follow-up were considered to have undefined outcome.
Data Source and Measurement A questionnaire was developed locally, piloted, refined, and then used to interview patients and collect socio-demographic data concerning household income and expenses throughout TB illness ( Questionnaire S1 ). Interviews were conducted at baseline with both TB patients and controls. For patients, this baseline interview occurred prior to or at the time that treatment commenced. The baseline interview (but not subsequent interviews) included detailed assessment of household assets ownership, access to basic services, and education level. Patients were subsequently interviewed after 2, 4, 6, 8, 12, 16, 20, and 24 wk of treatment. At the baseline and all subsequent interviews, questions characterised earnings, income, expenses, employment (paid or unpaid), number of days unable to work due to illness, additional household food expenditure due to TB illness, and crowding. Household debts were assessed at recruitment and subsequently at 24 wk of treatment.
As in previous research, TB-related costs were categorised as “direct expensesterm·key” [6] , [9] , [30] , [33] and “lost incometerm” [6] , [33] , [34] incurred since the previous interview. All costs and incomes were quantified in cash amounts in Peruvian Soles (PEN) (US$1 on average equivalent to 2.9 PEN during the study period). Inflation and especially exchange rates varied considerably during the study period, so actual costs were reported without adjustments in order to be more informative to users including policy-makers. Table S6 shows annual inflation in Peruterm·key and average annual exchange rates for 2002–2009. To further facilitate interpretation internationally, costs were also expressed as the proportion of the average monthly income of all patient households in the cohort (termed “monthly incomes”). Also, to assess impact on the patient households, costs were calculated as the proportion of the same household's annual income.
“Direct (out-of-pocket) expenses” included direct medical expenses (medical examinations and prescribed medicines) and direct non-medical expenses (natural non-prescribed remedies, TB-care-related transport, extra food, and other miscellaneous expenses). “Lost incometerm” (indirect expensesterm) was the income the patient estimated that the household lost due to TB illness or TB-related time off work (such as attending clinics) since the previous interview, measured in Peruvian Soles. Number of days of work lost due to TB illness could not be used to directly calculate lost incometerm because salaried employment with fixed rates of remuneration was uncommon in this setting. “Total expensesterm” were direct expensesterm·key plus lost incometerm. “Earnings” were defined as the monthly money actually received by the household, and “income” was defined as the monthly money earned by the household plus lost incometerm. Household debts at recruitment and total household debts (sum of debts at recruitment plus debts at 24 wk of recruitment) included both formal debts (e.g., bank loans) and informal debts (e.g., money borrowed from friends and family).
For all participants, height and weight were measured and body mass index (BMI) was calculated. Povertyterm was measured using a composite household povertyterm index in arbitrary units derived by principal component analysis from 13 variables, as previously described [35] .
A threshold for catastrophic coststerm·dom was calculated by plotting the sensitivity, specificity, and population attributable fractionterm for adverse TB outcometerm against total household expenses as a proportion of annual income. In order to assess the strength of this new definition of catastrophic coststerm·dom and in accordance with relevant recent studies [16] , [20] , a sensitivity analysis was also performed comparing the association of other existing catastrophic coststerm·dom thresholds (including total expensesterm equal to or greater than 10%, 15%, or 25% of annual income) with adverse TB outcometerm.
Data Analysis Continuous data were summarised by their arithmetic means and their 95% confidence intervals (CIs) whether the data were Gaussian or non-Gaussian because this approach is considered to be robust for health economics data analysis [7] , [36] , [37] . Furthermore, because of the skewed nature of some expenditure data, most median values were zero or close to zero, limiting the descriptive usefulness of presenting median values. Any direct expensesterm·key, lost incometerm, or annual income recorded as “zero” or missing was replaced with 0.5 PEN per day (i.e., the midpoint of zero and the lowest unit of measurement, 1 PEN). Means were compared with the Student's t -test. Categorical data were summarised as proportions with 95% CIs and were compared with the z -test of proportions. Univariable regression analyses examining differences between patients with MDRterm·dom and non-MDR TB and controls were adjusted for sex because of under-recruitment of male controls due to their availability.
The association between catastrophic coststerm·dom and adverse TB outcometerm was explored through both univariable and multivariable analysis to determine odds ratios (ORs). The likelihood ratio test was used to test for trend and interaction between variables. Non-Gaussian continuous variables such as total costs as a proportion of annual income were transformed to their base-10 logarithm for regression analysis. All the variables associated ( p <0.15) with adverse TB outcometerm in univariable analysis and all predetermined presumed confounding variables (age, povertyterm score, previous TB episode, symptom duration, and current MDRterm·dom TB) were concurrently included in a multivariable model [38] .
Population attributable fractionsterm were calculated using the Stata program “aflogit” function, which computes population attributable fractionterm estimates while adjusting for the reciprocal confounding effect of covariates on the association of interest. The population attributable fractionterm of an exposure was interpreted as the proportion of adverse TB outcomesterm that would be averted by eliminating that exposure, both unadjusted and adjusted for known confounding factors. All p -values were two-sided, and statistical analyses were performed using the Stata program (version 10, StataCorp).
Results Participants During the study period, the Peruvian national TB program within the study site of Ventanilla registered 1,014 patients. We located 99% of these registered TB patients, of whom 95% ( n = 966) met the inclusion criterion. Of these eligible patients, 1% ( n = 10) declined, and 8% ( n = 80) were excluded because they completed fewer than half of our planned research interviews; data are presented for the remaining 91% ( n = 876). 11% ( n = 93) of patients recruited had MDRterm·dom TB. 487 controls were also recruited and had only a baseline interview. The characteristics of the study population are summarised in Table 1 .
Study population baseline data.
Descriptive Data TB patients were more likely than controls to be younger (mean age 31 [95% CI = 18–44] versus 34 [95% CI = 30–48] y old, p = 0.001), to be male (59% [95% CI = 5%5–62%] versus 37% [95% CI = 33%–41%] male, p <0.001), to have a lower BMI (21 [95% CI = 21–22] versus 26 [95% CI = 25–26] kg/m 2 , p <0.001), to have lower earnings (510 [95% CI = 481–539] versus 651 [95% CI = 595–707] PEN, p <0.001), to not be in paid work at recruitment (81% [95% CI = 79%–84%] versus 63% [95% CI = 56%–69%], p <0.001), and to have had a previous TB episode (18% [95% CI = 15%–20%] versus 5.4% [95% CI = 3.3%–7.4%] of individuals, p <0.001). Patients with MDRterm·dom TB were more likely than patients with non-MDR TB to have had a previous TB episode (40% [95% CI = 30%–50%] versus 15% [95% CI = 13%–18%] of individuals, p <0.001), to have longer pre-treatment symptom duration (83 [95% CI = 0–192] versus 52 [95% CI = 0–118] d, p <0.001), to not be in paid work (90% [95% CI = 84%–96%] versus 80% [95% CI = 77%–83%] of individuals, p <0.03), and to have had more days not working pre-treatment due to TB-related illness (29 [95% CI = 0–81] versus 18 [95% CI = 0–47] d, p = 0.004).
Patients earned more per month pre-treatment than during treatment (510 [95% CI = 481–539] versus 434 [95% CI = 415–453] PEN, p <0.001; Table 2 ) or the continuation treatment phase (454 [95% CI = 431–477] PEN, p <0.001), and earned least during the intensive treatment phase (379 [95% CI = 358–400] PEN, p <0.001). During all treatment phases, patients with MDRterm·dom TB tended to earn less than patients with non-MDR TB ( Table 2 ). Household debts at recruitment were greater in controls than in patients with TB (812 [95% CI = 507–1117] versus 383 [95% CI = 292–474] PEN, p = 0.004), but there was no difference between the household debts of patients with MDRterm·dom and non-MDR TB (497 [95% CI = 381–613] versus 511 [95% CI = 482–540] PEN, p = 0.8). Household debts decreased from 383 (95% CI = 292–474) PEN at recruitment to 296 (95% CI = 176–414) PEN at 24 wk of treatment. Households with total debt above the cohort median were more likely to incur catastrophic coststerm·dom (OR 1.58 [95% CI = 1.17–2.14], p = 0.003). Households with above-cohort-median increase in debt from recruitment to 24 wk were also more likely to incur catastrophic coststerm·dom (OR 1.74 [95% CI = 1.10–2.77], p = 0.003).
Comparison of mean monthly earnings of patient households by treatment stage.
Outcome Data 725 (83%) patients had a defined TB outcome at follow-up. Of these patients, 166 (23%) had an adverse TB outcometerm, 40% ( n = 67) due to treatment abandonment, 22% ( n = 36) due to treatment failure, 12% ( n = 15) due to death during treatment, and 26% ( n = 48) due to TB recurrence.
Costs Data Direct expensesterm·key and lost incometerm Direct expensesterm·key and lost incometerm are summarised in Figures 1 and S1 . Throughout the entire illness, the proportions of direct expensesterm·key that were medical and non-medical were similar (49% [95% CI = 43%–55%] versus 51% [95% CI = 47%–54%] of total direct expensesterm·key, p = 0.7). Medical expenses were greatest pre-treatment: during this period, medical expenses were higher than non-medical expenses and constituted almost two-thirds of overall direct expensesterm·key (0.20 [95% CI = 0.18–0.22] versus 0.13 [95% CI = 0.11–0.15] monthly incomes, p <0.001). Conversely, during treatment, non-medical expenses were higher than medical expenses and constituted approximately two-thirds of overall direct expensesterm·key (0.22 [95% CI = 0.20–0.24] versus 0.14 [95% CI = 0.12–0.16] monthly incomes, p <0.001). Direct expensesterm·key were higher pre-treatment than during treatment (0.52 [95% CI = 0.46–0.59] versus 0.41 [95% CI = 0.37–0.44] monthly incomes, p <0.001), whereas lost household income was lower pre-treatment than during treatment (0.60 [95% CI = 0.50–0.69] versus 0.75 [95% CI = 0.68–0.82] monthly incomes, p <0.005). Lost household income was higher than direct expensesterm·key throughout all treatment phases, with the greatest difference during the intensive treatment phase (69% lost incometerm [95% CI = 61%–77%]; Figures 1 and S1 ).
Lost incometerm, direct expensesterm·key, and total expensesterm by treatment stage in mean Peruvian Soles (PEN) and as a proportion of mean monthly household income. The top row of data in the table below the bar graph shows lost incometerm in mean Peruvian Soles and, in parentheses, as a percentage of total expensesterm. The next six rows show direct expensesterm·key in mean Peruvian Soles and, in parentheses, as a percentage of total direct expensesterm·key. Medical expenses are defined as the sum of direct expensesterm·key for medicines (blue bar) and clinical exams (dark blue bar); non-medical expenses are defined as the sum of direct expensesterm·key for natural remedies, TB-care-related transport, extra food, and other TB-related expenses. The lowermost two rows show total direct expensesterm·key (i.e., sum of medicines, clinical exams, natural remedies, transport, extra food, and other expenses) and total expensesterm in mean Peruvian Soles and, in parentheses, as a percentage of total expensesterm. p -Values represent the difference between treatment stages by Student's t -test. 23/876 (2.6%) of the TB patient cohort had direct expensesterm·key of 0 PEN, and 14/876 (1.6%) had total expensesterm of 0 PEN, and thus these zero values were replaced with 0.5 PEN per day. A line chart representation of this graph is available in Figure S1 .
Total expensesterm In addition to direct expensesterm·key and lost incometerm, total expensesterm are summarised in Figures 1 and S1 . Total expensesterm were similar pre-treatment and during treatment (1.1 [95% CI = 1.0–1.2] versus 1.2 [95% CI = 1.1–1.2] monthly incomes, p = 0.6). Total expensesterm (1.12 [95% CI = 0.99–1.25] versus 0.62 [95% CI = 0.56–0.68] monthly incomes, p <0.001), direct expensesterm·key (0.52 [95% CI = 0.46–0.59] versus 0.19 [95% CI = 0.16–0.22] monthly incomes, p <0.001), and lost incometerm (0.6 [95% CI = 0.50–0.69] versus 0.43 [95% CI = 0.38–0.48] monthly incomes, p = 0.001) were significantly higher pre-treatment than during the intensive treatment phase. Total expensesterm (0.62 [95% CI = 0.56–0.68] versus 0.54 [95% CI = 0.49–0.59] monthly incomes, p = 0.01) and lost incometerm (0.43 [95% CI = 0.38–0.48 versus 0.32 [95% CI = 0.28–0.36] monthly incomes, p <0.001) were higher in the intensive than the continuation treatment phase, but there was no difference in direct expensesterm·key between these treatment phases (0.19 [95% CI = 0.16–0.22] versus 0.22 [95% CI = 0.20–0.23] monthly incomes, p = 0.07). When total expensesterm were examined per month, monthly total expensesterm for the intensive treatment phase were approximately double those of the continuation treatment phase.
Povertyterm and expenses TB patients were poorer than controls (58% [95% CI = 55%–61%] versus 51% [95% CI = 47%–56%] above control mean, p <0.02; Table 1 ). In poorer households, direct expensesterm·key were lower (mean direct expensesterm·key of poorest households 330 [95% CI = 287–373], poor households 418 [95% CI = 351–485], and least-poor households 435 [95% CI = 380–490] PEN, p <0.001; Figure 2 ), but total expensesterm made up a greater proportion of the same household's annual income (poorest households 48% [95% CI = 35%–50%], poor households 47% [95% CI = 24%–70%], and least-poor households 27% [95% CI = 20%–34%], p <0.001; Figure 2 ).
Expenses and economic burden of TB illness across povertyterm terciles. (A) Total expensesterm as proportion of annual income. (B) Direct expensesterm·key. p -Values represent Pearson's coefficient of trend. Bars represent confidence intervals. The numbers in the three bars of (A) refer to the left-hand y -axis of total expensesterm as a proportion of annual household income. The numbers in the three bars of (B) refer to the right-hand y -axis of direct expensesterm·key in mean Peruvian Soles.
Main Findings Catastrophic coststerm·dom A threshold of total expensesterm ≥20% of annual household income was defined as catastrophic because this threshold had the highest sensitivity, specificity, and population attributable fractionterm for association with adverse outcome ( Figure 3 ). Catastrophic coststerm·dom were incurred by 345 households (39%). Incurring catastrophic coststerm·dom was independently associated with MDRterm·dom TB (OR 1.61 [95% CI = 0.98–2.64], p <0.06), more days not working pre-treatment (OR 1.00 [95% CI = 1.00–1.01], p = 0.03), greater debts at recruitment (OR 1.00 [95% CI = 1.00–1.00], p = 0.02), being male (OR 2.16 [95% CI = 1.57–2.96], p <0.001), being older (OR 1.01 [95% CI = 1.00–1.03], p = 0.02), being poorer (OR 1.25 [95% CI = 1.15–1.36], p <0.001) and not being in paid employment (OR 1.86 [95% CI = 1.23–2.79], p = 0.003; Table 3 ). Households of patients who had MDRterm·dom TB were more likely to incur catastrophic coststerm·dom than households of patients with non-MDR TB (54% [95% CI = 43%–64%] versus 38% [95% CI = 34%–41%], p <0.003; Figure 4 ). When the catastrophic coststerm·dom multivariable regression analyses were repeated with total costs as a proportion of annual income analysed as a continuous outcome variable (instead of a dichotomous variable: above versus below a threshold indicating catastrophic coststerm·dom), the results and patterns of significance were similar ( Table S1 ).
Sensitivity, specificity, and univariable population attributable fractionterm of the association of total expensesterm as a proportion of annual income with adverse TB outcometerm. Total household TB-associated costs were defined as catastrophic when they met or exceeded 20% of household annual income because this threshold had the highest sensitivity, specificity and population attributable fractionterm for association with adverse outcome.
Patient households incurring catastrophic coststerm·dom by TB resistance profile and adverse TB outcometerm. Error bars represent 95% confidence intervals. p -Values represent association in univariable logistic regression.
Factors associated with incurring catastrophic coststerm·dom.
Catastrophic coststerm·dom and adverse TB outcometerm Of the 725 patients with both TB outcome and catastrophic coststerm·dom data, 166 (23%) had an adverse TB outcometerm. In multivariable regression analysis, having MDRterm·dom TB was most strongly associated with adverse TB outcometerm (OR 8.4 [95% CI = 4.7–15], p <0.001). Having had previous TB (OR 2.1 [95% CI = 1.3–3.5], p = 0.005), having more days not working due to illness prior to TB diagnosis (OR 1.01 [95% CI = 1.00–1.01], p = 0.02), and incurring catastrophic coststerm·dom (OR 1.7 [95% CI = 1.1–2.6], p = 0.01; Table 4 and Figure 5 ) were also independently associated with adverse TB outcometerm. When the adverse TB outcometerm multivariable regression analyses were repeated with total costs as a proportion of annual income analysed as a continuous outcome variable (instead of a dichotomous variable: above versus below a threshold indicating catastrophic coststerm·dom), the results and patterns of significance were similar ( Table S2 ). The likelihood ratio test did not reveal any interaction between having MDRterm·dom TB and incurring catastrophic coststerm·dom (whether analysed using quantiles of costs as a continuous variable or using our catastrophic coststerm·dom threshold).
Percentage of patients experiencing an adverse TB outcometerm analysed by povertyterm, education level, symptom duration, time too unwell to work, catastrophic coststerm·dom, previous TB, and resistance profile. Error bars represent 95% confidence intervals. p -Values correspond to the association of each variable with adverse TB outcometerm in univariable logistic regression, except for povertyterm and symptom duration, which were analysed as continuous variables. In multivariable regression analysis, the following variables remained independently associated with adverse TB outcometerm: time too unwell to work ( p = 0.02), catastrophic coststerm·dom ( p = 0.003), having had a previous episode of TB ( p = 0.004), and currently having MDRterm·dom TB ( p <0.0001).
Univariable and multivariable logistic regression of factors associated with adverse TB outcometerm.
Population attributable fractionterm The unadjusted population attributable fractionterm of adverse TB outcomesterm explained by catastrophic coststerm·dom was 26% (95% CI = 14%–36%), similar to that of MDRterm·dom TB (23% [95% CI = 17%–28%]). When catastrophic coststerm·dom, MDRterm·dom TB, and previous TB episode were included in the multivariable regression model, the adjusted population attributable fractionterm of adverse TB outcomesterm explained by catastrophic coststerm·dom and MDRterm·dom TB was similar: 18% (95% CI = 6.9%–28%) for catastrophic coststerm·dom and 20% (95% CI = 14%–25%) for MDRterm·dom TB.
Sensitivity analysis of other catastrophic coststerm·dom thresholds Using a threshold of total costs of 10% or more of annual income, 578 patients (66%) incurred catastrophic coststerm·dom. When this threshold was increased to total costs of 15% or more of annual income, 457 patients (52%) incurred catastrophic coststerm·dom. Finally, at a threshold of total costs of 25% or more of annual income, 281 patients (32%) incurred catastrophic coststerm·dom. When these thresholds were included in the multivariable regression models, it was found that catastrophic coststerm·dom at a threshold of 10% or more, or 15% or more, of annual income were not independently associated with adverse TB outcometerm ( Tables S3 and S4 ). Conversely, catastrophic coststerm·dom at a threshold of 25% or more of annual income were independently associated with adverse TB outcometerm ( Table S5 ).
Discussion In this prospective cohort study of TB patients in impoverished Peruvian shantytowns, accessing free TB care was expensive for poor TB patients, especially those with MDRterm·dom TB. Our novel findings also define an evidence-based threshold for catastrophic coststerm·dom of total costs greater than or equal to 20% of annual income for TB-affected households. Moreover, an additional sensitivity analysis revealed that other recognised catastrophic coststerm·dom thresholds (such as expenses equal to or greater than 10% or 15% of annual income) did not identify any association between catastrophic coststerm·dom and adverse TB outcometerm in this setting. Our new definition is innovative in demonstrating the strong association between this 20% catastrophic coststerm·dom threshold and the increased chances of adverse TB outcometerm, independent of MDRterm·dom TB status. Our population attributable fractionterm analysis supports this observation by indicating that a similar proportion of adverse TB outcomesterm may be averted by eliminating catastrophic coststerm·dom or MDRterm·dom TB from the study population. Overall, our findings suggest that the costs imposed on TB-affected households are not only financially but also clinically relevant, and highlight the importance of household povertyterm and catastrophic TB-related costs in relation to TB epidemiology, outcome, and control.
Social determinants are important in the causal pathway of TB disease [39] . Indeed, reducing povertyterm, advocating improved equity of access and universal healthcare, and eliminating catastrophic coststerm·dom in TB-affected households are key components of the World Health Organization's post-2015 global TB strategy [13] , [14] , [26] , [40] . Our results demonstrate that both drug-susceptible and MDRterm·dom TB predominantly affect the poor [3] , [4] , [41] , [42] . Our povertyterm score used time-stable variables such as household assets, education level, and housing [43] ; thus, povertyterm can be assumed to have preceded TB disease. The extent and effect of povertyterm may be underestimated in our study given that geographical and socioeconomic barriers and stigma may particularly preclude the poorest people from seeking healthcare [44] .
As in previous studies [8] , [9] , [11] , poorer people incurred the most catastrophic coststerm·dom, which they were less able to afford, probably causing further impoverishment [7] , [44] . In agreement with other findings, our results indicate that costs as a proportion of the same household's income indicate economic challenge better than actual monetary expenditures: this finding highlights the “medical povertyterm trap”, that as incomes decrease, proportional costs increase [6] , [45] . In addition to exemplifying this medical povertyterm trap, our results also show that being poorer was independently associated with incurring catastrophic coststerm·dom. Catastrophic coststerm·dom in poor households can lead to financial shock: families reducing consumption below minimum needs, selling assets, and taking children out of education. These actions may in turn increase stigmatisation [17] , [18] , [46] , [47] . Moreover, TB principally affects the most economically productive age group, and patient and household income decreases post-diagnosis [6] and may not return to pre-diagnosis levels. Our study adds a new dimension to the social protectionterm·key TB literature by showing how catastrophic coststerm·dom can also have significant clinical implications: loss of income and higher hidden costs have previously been associated with poor treatment adherence and high dropout rates in TB patients [9] , [44] , [48] , but their independent effects on long-term TB outcome have not been previously characterised. We hypothesize that the relationship we found between catastrophic coststerm·dom and adverse TB outcometerm may relate to a number of factors along the causal pathway from TB susceptibility to illness to recurrence, including inadequate nutrition due to lower food spending, more severe disease (both MDRterm·dom and non-MDR TB), and barriers to cure due to the disproportionate hidden costs associated with adherence to and completion of treatment. These adverse TB outcomesterm associated with catastrophic coststerm·dom may increase TB and MDR-TB transmission, especially in poorer households. Thus, catastrophic coststerm·dom may worsen TB control.
Regardless of the mechanisms mediating the association between catastrophic coststerm·dom and adverse TB outcometerm, the policy implications of our findings are clear: future TB prevention strategies should incorporate social protectionterm·key to mitigate decreased economic production, loss of employment, and TB-associated povertyterm, and to reduce the clinical vulnerability of TB patients. These findings highlight the potential role of social protectionterm·key not just as a poverty-reduction strategy, but also as a tool to improve disease control and, ultimately, health [23] , [24] .
Our previous social protectionterm·key intervention project, Innovative Socioeconomic Interventions Against Tuberculosisterm·key (ISIAT), provided evidence that in Peruterm·key multidisciplinary social protectionterm·key intervention can improve adherence and completion of TB treatment and prophylaxis [35] . Social protectionterm·key interventions targeting disadvantaged and vulnerable populations have shown much promise in Latin America [49] , for example, conditional cash transfer projects such as the Programa de Educación, Salud y Alimentación (Progresa) in Mexico [50] . In order for these programs to be adopted on a larger scale and reduce the social health gradient, they require a rigorous evidence base that is currently lacking. Our present study suggests that by reducing catastrophic coststerm·dom, social protectionterm·key has the potential to protect families from TB illness and deepening povertyterm [17] , [51] .
Despite TB treatment being free of direct charges in Peruterm·key, the overall costs in TB-affected households were high and were similar pre-treatment and during treatment. Regardless of MDRterm·dom TB status, higher total expensesterm were incurred in households where the patient had longer symptom duration prior to diagnosis, consistent with the known association between increased expenses and diagnostic delay [40] . A strength of our study is that it analysed patients with both MDRterm·dom and non-MDR TB and found that expenses as a proportion of household annual income were significantly higher in patients with MDRterm·dom TB, as has been noted in other Latin American countries [52] , [53] .
Medical expenses made up the largest proportion of pre-treatment direct expensesterm·key, probably because TB care was provided free of charge only when TB was being tested for or after TB was diagnosed. Consequently, formal medical care for the presenting illness was often expensive for the patient prior to TB being suspected and/or diagnosed [43] , [54] . We found no evidence of healthcare providers requesting “informal” or “under the table” payments, costs that have been reported in other countries [43] .
Extending findings from previous studies [6] , [9] , [11] , lost incometerm formed the majority of the economic burden of total expensesterm, and TB patients, especially those with MDRterm·dom TB, were more likely than controls to have a lower income, to be without paid work, and to miss work due to TB-related illness. In addition, having more days too unwell to work due to TB illness was independently associated with having an adverse TB outcometerm. These results suggest that the socioeconomic and employment situation of TB patients is often precarious, and this may negatively impact their health, as has been found in another study [55] . Indeed, our finding that having more days too unwell to work was independently associated with both incurring catastrophic coststerm·dom and having an adverse TB outcometerm suggests that TB illness in such patients may be more severe or advanced, and financial shock more likely.
Some definitions of catastrophic coststerm·dom incorporate signs of financial shock, when a household is forced to employ coping mechanisms such as sacrificing basic needs, selling assets, selling household items, removing children from education, and incurring formal or informal debt [17] , [18] , [46] , [47] , [56] , [57] . Others have defined costs as catastrophic when they exceed 10%–40% of annual household or individual income [7] , [15] , [56] , [58] or 40% or more of a household's “capacity to pay” (the effective income for non-food spending [21] , [22] , [46] , [59] – [61] ), but this approach may be too narrow and potentially misleading to policy-makers because it overlooks lost incometerm [23] , [24] . A strength of the threshold of catastrophic coststerm·dom that our results defined is that it includes not only out-of-pocket direct expensesterm·key but also lost incometerm and that it is proven to be clinically relevant. Specifically, our definition was calculated from serial, prospective data [15] , [18] of household expenses, actual household income [7] , [58] , and long-term TB outcome of a cohort of TB patients in impoverished Peruvian shantytowns. It has been estimated that 4% of households in Peruterm·key incur catastrophic health expenditure when aiming to meet overall health needs [62] . Rates of catastrophic health expenditure in our cohort were much higher than those of the general population. This may be due to TB-affected households being poorer or the TB treatment model in Peruterm·key having greater hidden costs for TB patients, or that we included lost incometerm to calculate catastrophic coststerm·dom, whereas only direct expensesterm·key were used in some other studies [62] .
The sensitivity analysis we performed showed that the proportion of patient households incurring catastrophic coststerm·dom was similar to the proportion found in other studies that used different thresholds: at a threshold of total costs of 10% or more of annual income, 65% of our cohort incurred catastrophic coststerm·dom, compared to 66%–75% in related studies from sub-Saharan Africa [12] , [16] ; at thresholds of total costs of 15% and 25% or more of annual income, 52% and 32% of our cohort, respectively, had catastrophic coststerm·dom, compared to 68% and 48%, respectively, in a cohort from sub-Saharan Africa [12] . More importantly, the sensitivity analysis also showed that thresholds of total costs of 10% and 15% or more of annual income were not independently associated with adverse TB outcometerm in this Peruvian shantytown setting. Our results demonstrate that these previously published arbitrary thresholds for catastrophic coststerm·dom that were defined without patient follow-up were not associated with adverse TB outcometerm for TB patients in our setting. Thus, our findings provide, to our knowledge, the first evidence-based threshold for clinically relevant catastrophic coststerm·dom, and demonstrate a methodology to assess the generalizability of this threshold in other settings.
This study has several limitations. First, cases and controls were not matched in this study because controls were specifically included to provide an estimate of typical income and expenditure in this community, to be compared with TB patients at baseline. Matching would have impaired this comparison. A baseline difference was noted in debts at recruitment, a proxy for “dis-saving”. Poorer households have diminished access to establishments that offer formal loans (e.g., banks) because of their uncertain repayment capacity and/or lack of requisites such as a national identity card [35] . However, even if controls had been matched to cases, controls may still have had higher debt than patients because some patient households were not eligible for some loans due to serious ill health or if they were extremely poor [35] . Apart from debt, no other data were collected on specific “dis-saving” coping mechanisms. However, although selling household items may have been overlooked, taking children out of education and selling livestock are unlikely to occur in the peri-urban non-agrarian communities that made up our study cohort. Second, the data available did not allow assessment of an existing WHO definition of catastrophic coststerm·dom (40% or more of a household's capacity to pay [63] ) against which other studies have compared their findings [12] . Third, we may have underestimated the financial effects of MDRterm·dom TB because our questionnaires quantifying costs continued for only 6 mo, whereas patients with MDRterm·dom TB are usually treated for 18 mo or more. We decided a priori to analyse the catastrophic coststerm·dom of both MDRterm·dom and non-MDR patients together, given their equal follow-up and the small number of MDRterm·dom TB patients. Finally, our research demonstrates a new methodology that should be repeated in other settings to assess the external validity of our findings.
Despite free TB care, having TB disease was expensive for TB patients living in a shantytown in Peruterm·key. Higher relative costs were associated with greater likelihood of adverse TB outcometerm. Having MDRterm·dom TB and incurring catastrophic coststerm·dom were independently associated with adverse TB outcometerm, with a similar adjusted population attributable fractionterm for adverse TB outcometerm. Thus, catastrophic coststerm·dom were an indicator of both financial and clinical vulnerability, and households affected by TB would benefit from assessment to identify those at highest risk of incurring catastrophic coststerm·dom. Mitigating catastrophic coststerm·dom through targeted social protectionterm·key interventions as well as prompt diagnosis and appropriate treatment of MDRterm·dom TB deserve attention in TB control programs. In conclusion, control interventions must consider TB as an infectious and socioeconomic problem and address both the clinical and financial aspects of this public health challenge.
Supporting Information Lost incometerm, direct expensesterm·key, and total expensesterm by treatment stage in mean Peruvian Soles and as a proportion of mean monthly household income. A line chart representation of the data presented in Figure 1 .
(TIF) Click here for additional data file.
Factors associated with total costs as a proportion of annual income. Total costs as a proportion of annual income had a non-Gaussian distribution, so this variable was transformed to its base-10 logarithm for regression analysis. Factors associated ( p <0.15) with increasing costs in univariable linear regression were included in the multivariable analysis. 95% confidence intervals are shown in parentheses. All patients ( n = 876) had data available and were included in the univariable and multivariable linear regression analyses.
(DOC) Click here for additional data file.
Univariable and multivariable logistic regression of factors (including costs as a continuous variable) associated with adverse outcome. Adverse outcome is defined as death during treatment, treatment failure or abandonment, or recurrence of TB within 30 mo of starting treatment. Total costs as a proportion of annual income had a non-Gaussian distribution so this variable was transformed to its base-10 logarithm for regression analysis. Factors associated ( p <0.15) with adverse outcome in univariable logistic regression were included in the multivariable logistic regression analysis. 725/876 (83%) of patients had outcome data available and were included in the univariable and multivariable logistic regression analyses. This table differs from Table 4 in that total costs as a proportion of annual income is analysed as a continuous variable instead of as a dichotomous variable (catastrophic versus non-catastrophic costs).
(DOC) Click here for additional data file.
Univariable and multivariable logistic regression of factors (including 10% threshold for catastrophic coststerm·dom) associated with adverse outcome. Adverse outcome is defined as death during treatment, treatment failure or abandonment, or recurrence of TB within 30 mo of starting treatment. Factors associated ( p <0.15) with adverse outcome in univariable logistic regression were included in the multivariable logistic regression analysis. 725/876 (83%) of patients had outcome data available and entered the univariable and multivariable logistic regression analyses. In contrast to Table 4 , in this table total costs ≥10% of annual income was used as the threshold for catastrophic coststerm·dom.
(DOC) Click here for additional data file.
Univariable and multivariable logistic regression of factors (including 15% threshold for catastrophic coststerm·dom) associated with adverse outcome. Adverse outcome is defined as death during treatment, treatment failure or abandonment, or recurrence of TB within 30 mo of starting treatment. Factors associated ( p <0.15) with adverse outcome in univariable logistic regression were included in the multivariable logistic regression analysis. 725/876 (83%) of patients had outcome data available and entered the univariable and multivariable logistic regression analyses. In contrast to Table 4 , in this table total costs ≥15% of annual income was used as the threshold for catastrophic coststerm·dom.
(DOC) Click here for additional data file.
Univariable and multivariable logistic regression of factors (including 25% threshold for catastrophic coststerm·dom) associated with adverse outcome. Adverse outcome is defined as death during treatment, treatment failure or abandonment, or recurrence of TB within 30 mo of starting treatment. Factors associated ( p <0.15) with adverse outcome in univariable logistic regression were included in the multivariable logistic regression analysis. 725/876 (83%) of patients had outcome data available and were included in the univariable and multivariable logistic regression analyses. In contrast to Table 4 , in this table total costs ≥25% of annual income was used as the threshold for catastrophic coststerm·dom.
(DOC) Click here for additional data file.
Annual inflation rate of the Peruvian Sol and exchange rate of the Peruvian Sol. Source: l PEN to the US dollar, 2002–2009 [64] .
(DOC) Click here for additional data file.
Socioeconomic section of initial and follow-up questionnaires.
(DOC) Click here for additional data file.