Fig 1.
Trends in Xpert test coverage and modeled prevalence of rifampicin resistance among notified TB cases.
(A) The proportion of new and previously treated TB cases tested with Xpert between 2014–2023 with a conclusive rifampicin resistance result, defined as being either susceptible (RR-TB negative) or resistant (RR-TB positive). This excludes observations where patients were tested with Xpert, but their result was labeled as indeterminant, not recorded, or was not positive for TB with Xpert despite receiving a TB diagnosis. (B) Lines reflect the modeled prevalence of rifampicin resistance among all notified TB cases for a given quarter by case type for 2017–2023. Points reflect the prevalence of rifampicin resistance calculated using the naïve approach, defined as the share of all conclusive Xpert results with rifampicin resistance. Point size indicates the number of notified TB cases with a recorded Xpert rifampicin test result.
Table 1.
Descriptive characteristics of notified TB cases (2017–2023).
Fig 2.
Extent to which naive Xpert results underestimates the prevalence of rifampicin resistance among notified TB cases (2017–2023).
Each line reflects the bias in observed data as a function of the ratio of modeled to naïve prevalence among notified TB cases per 100,000 person-years. Prevalence of rifampicin resistance is calculated as either the modeled number of RR-TB cases (modeled) or the observed number of RR-TB cases scaled by the fraction with conclusive Xpert rifampicin results (naïve), divided by the national population in 2010. Numerators are calculated quarterly and are scaled to obtain person-years. This should be interpreted as how much higher estimates of rifampicin resistance prevalence using our approach are relative to naïve estimates. The dashed line at 1 indicates where there would be no bias between modeled and naïve estimates. 95% uncertainty intervals are shaded.
Fig 3.
Rifampicin resistance prevalence among notified TB cases and total incidence comparing modeled to WHO estimates (2017–2023).
(A) and (B) Modeled prevalence of rifampicin resistance by case type and 95% uncertainty intervals that are shaded. (C) Modeled total number of RR-TB cases among notified TB cases per 100,000 person-years and corresponding 95% uncertainty intervals (“Modeled”). It also presents total incidence after adjusting modeled estimates by Brazil’s case detection rate (CDR) to account for underreporting of TB cases (“CDR-inflated”) [16]. Naïve refers to the number of RR-TB cases calculated from Xpert MTB/RIF only (“Naïve (Xpert)”) and from all DST results (“Naïve (Xpert + DST)”) test results, scaled by the fraction of notified TB cases that were tested. DST results are included if resistance to at least rifampicin is indicated. All three panels are overlaid by the corresponding estimates and 95% uncertainty intervals from the World Health Organization’s Global Tuberculosis [13]. Since CDR and RR-TB estimates from WHO are only available through 2022, 2023 estimates were carried over from 2022.
Fig 4.
Modeled levels of prevalence of rifampicin resistance by state, 2023.
Each panel presents a map of the modeled prevalence of rifampicin resistance among notified TB cases by state (left) and point estimates with 95% uncertainty intervals by each state and region (right). Two-digit codes associated with each state are listed alongside the state name on the right. Estimates have been plotted in R using the basemap shapefiles provided by the Brazilian Institute of Demography and Statistics (IBGE): https://geoftp.ibge.gov.br/organizacao_do_territorio/malhas_territoriais/malhas_municipais/municipio_2020/Brasil/BR/BR_UF_2020.zip.
Fig 5.
Selected state-level trends in RR-TB prevalence among notified TB cases per 100,000 person-years by case type (2017–2023).
Selected states are those with the highest number of RR-TB cases among notified TB cases per 100,000 person-years and who tested at least 30% of notified TB cases in 2023. “Modeled” reflects modeled estimates and shaded 95% uncertainty intervals. “Naïve” estimates are only among Xpert MTB/RIF test results. States are ordered highest to lowest based on the number of RR-TB cases among notified TB cases per 100,000 person-years in 2023.
Fig 6.
Alternative model specifications.
The above compares modeled results from several alternative models by time period (A) and specification (B). (A) Extends the primary specification from 2017–2023 to include the early period of Xpert MTB/RIF implementation from 2014–2016. Across both panels, points reflect the naively estimated prevalence of rifampicin resistance—proportion of conclusive Xpert tests that are resistant. Point size indicates the number of cases with a conclusive Xpert test result. Lines reflect the modeled prevalence of rifampicin resistance among all notified TB cases for a given quarter, where line type corresponds to an alternative model. (A) Results from two alternative model specifications. The first model controls for additional patient covariates, including educational attainment, diabetes, drug consumption, tobacco consumption, alcohol consumption, whether they are experiencing homelessness, whether they are incarcerated, their immigration status, and race (“Additional Patient Covariates”). The second model fits smooth interactions between the original set of patient covariates in the model with time to determine whether there are any changes in selection over time (“Time-varying Selection”). These are compared to output from the primary specification (“Reference”).