This retrospective ecological panel study included country–year observations for Liberia and Sierra Leone from 2000 to 2022. There were 46 observations in total and 14 for the 2016–2022 subset.

Liberia and Sierra Leone served as the study settings. The period was 2000–2022, with a focused subset for 2016–2022 when TB-HIV testing coverage was available.

Data from all recorded country–year combinations for Liberia and Sierra Leone were utilised, with no sampling procedures applied. Analyses for each period were conducted using only complete cases.

TB indicators were obtained from the WHO Global Tuberculosis Database, including incidence rate, mortality not attributable to HIV, case notification rate, case detection rate, TB-HIV co-infection and TB-HIV testing coverage. Macroeconomic variables, specifically GDP per capita in constant 2015 US dollars and current health expenditure per capita in US dollars ($), were sourced from the World Bank World Development Indicators. Conflict event data were aggregated by country and year from Armed Conflict Location and Event Data (ACLED). Operational definitions, coding and measurement approaches, variable types, and data sources are detailed in Table 1. Tuberculosis-human immunodeficiency virus testing coverage data are available for 2016–2022.

Datasets were aligned using International Organization for Standardization (ISO3) country codes and calendar years before being combined into a balanced panel for both countries. Tuberculosis-human immunodeficiency virus testing coverage was calculated as the proportion of notified TB cases with a known HIV status, with values restricted to the 0% – 100% range following data quality assessments. Analyses for each period relied on complete cases only, and no data imputation was undertaken.

The primary analyses employed two-way fixed-effects models incorporating country and year indicators, estimated using the within transformation to account for unobserved country characteristics and year-specific influences common to both countries. Standard errors were adjusted for both serial and cross-sectional dependence using Driscoll-Kraay heteroskedasticity- and autocorrelation-consistent (HC1) estimators with a Bartlett kernel and a plug-in bandwidth. Given the small sample size (panel with only two countries, G = 2), statistical inference was limited. Therefore, emphasis was placed on effect sizes (coefficients) and their 95% Driscoll-Kraay confidence intervals (CIs). When p-values were presented in the main text, they were calculated using a conservative t reference distribution with degrees of freedom set to df = G – 1 = 1.

Models incorporating TB-HIV testing coverage were restricted to the 2016–2022 period, while analyses spanning the entire period (2000–2022) did not include this variable. Sensitivity analyses comprised linear mixed-effects regressions with a random intercept for country and fixed effects for year, as well as pooled ordinary least squares (OLS) models with country and year fixed effects and heteroskedasticity-robust (Heteroskedasticity–consistent standard error estimator, type 3 [HC3]) standard errors. As a result of the limited number of countries, cluster-robust CR2 standard errors were also assessed but yielded unstable estimates. As a result, HC3 standard errors are reported for the OLS robustness checks. All statistical analyses were performed using R version 4.4.2, employing plm (for two-way fixed effects and Driscoll-Kraay estimation via vcovSCC), lmtest (for inference), sandwich (for HC1 and HC3), lme4 and lmerTest packages. In Online Appendix 1 Table 1-A1–Table 4-A1, p-values are based on the asymptotic normal reference, and Driscoll-Kraay 95% CIs can be calculated as the coefficient estimate plus or minus 1.96 times the standard error reported.

Only publicly available, de-identified aggregate data were used in this research. Consistent with United States (US) federal regulations (45 CFR 46.104(d)(4)), the study did not constitute human or animal subjects research and therefore did not require approval from an institutional review board.

Liberia and Sierra Leone each provided data for 23 country-year periods, resulting in a total of 46 observations. The average TB incidence was 287.6 per 100 000 people in Liberia and 307.9 per 100 000 in Sierra Leone, with a combined mean of 297.7 per 100 000 (see Table 2). The mean TB-HIV co-infection rate was 22.1% for Liberia and 15.6% for Sierra Leone. Additional summary statistics are presented in Table 2.

Liberia’s estimated incidence rose from about 240 per 100 000 in 2000 to a peak near 308 in 2013 and then levelled through 2022. Sierra Leone started higher, increased modestly to about 318 in 2008, and declined to 283 by 2022. The post-2008 divergence suggests different epidemic trajectories and implementation timing of control efforts (Figure 1).

Across all study periods and estimation approaches, higher GDP per capita consistently corresponded with lower TB incidence and mortality. These results align with the effect estimates shown in Table 3 and Table 4 and Online Appendix 1 Table 4-A1.

The analysis restricted to the 2016–2022 period did not identify a statistically significant link between TB-HIV testing coverage and TB incidence. The positive coefficient suggests that expanding testing may increase the number of detected cases in the short term, even if actual transmission rates remain unchanged. This finding aligns with recommendations that testing must be integrated with same-day treatment initiation, ongoing care and dependable commodity and information systems to achieve population-level improvements.⁶ In fragile contexts, incomplete reporting and delays in data collection can further diminish the strength of observed associations between testing and incidence.⁸

Conflict events did not demonstrate a significant association with TB incidence in most models. However, the supplementary mortality analysis indicated a possible inverse relationship. This result should be interpreted with caution, as it may reflect limitations in reporting, challenges in event classification or changes in health service utilisation during periods of instability. Previous research during and after the 2014–2016 Ebola outbreak documents the indirect effects of such shocks on service delivery, access to care and routine surveillance in Liberia and Sierra Leone.^7,8,9

Programmes should align testing scale-up with same-day linkage and treatment capacity, and track time to treatment and treatment completion as routine indicators. Data systems should keep pace with testing by using stable patient identifiers and regular reconciliation across TB and HIV registers. Laboratories and essential commodities should be protected during budget cycles and shocks, so detection gains are followed by reductions in transmission. These actions operationalise integrated TB-HIV care and reinforce core health-security functions.^{2,5,6,7,10,11}

A health-security approach focuses on the core functions that keep populations protected during times of shock. The negative relationship between GDP per capita and TB burden reflects gains in these functions. Examples include reliable laboratories, stable supply chains, protected health workers and connected surveillance. Timely feedback is also crucial.^5,6,7,10 As these capacities improve, access to diagnosis and treatment becomes more dependable, and programme performance is easier to maintain.

Integrated TB-HIV services provide a practical framework for continuity. Shared workflows, common identifiers, harmonised registers, and coordinated commodity management help patients move from testing to treatment, even during disruptions. Whether facing outbreaks, staffing constraints or budget shortfalls, integration reduces drop-offs between case finding, linkage and treatment completion.^5,6,7,10 The same infrastructure also supports rapid recovery once operations stabilise, strengthening system resilience. As programmes rebuild, improved case finding and reporting may increase the number of detected cases before incidence declines.

A health-security approach anticipates this pattern. It pairs scale up with investments in treatment capacity, laboratory throughput, information systems and routine feedback cycles. These steps align with core system-strengthening priorities. Together, they increase the likelihood that improved detection leads to sustained reductions in TB incidence and mortality.^5,6,7,10

This analysis relies on publicly available surveillance and statistical model outputs that may carry measurement error, particularly when systems are disrupted. The number of country clusters is small, so inference is fragile, and emphasis is placed on effect sizes and CIs. Moreover, the study does not claim causal attribution. These limitations do not affect internal consistency, but they limit generalisability. Future research could apply this framework to additional fragile and conflict-affected contexts, examine variation in implementation of integrated TB-HIV services, and test whether improvements in linkage and data systems alter the relationship between expanded testing and reported incidence.