Longitudinal Symptom Transitions Predict Incident Tuberculosis Risk in Serial Active Case Finding Authors: Esther Jung, José Victor Bortolotto Bampi, Alessandra Moura da Silva, Yiran E Liu, Daniel Henrique Tsuha, Argita D Salindri, Andrea da Silva Santos, Roberto Dias de Oliveira, Julio Croda, Jason R Andrews Published: 2026 ABSTRACT AbstractTuberculosis (TB) screening often relies on cross-sectional symptom assessment. To determine the added value of longitudinal symptom monitoring, we conducted a prospective cohort study among 2282 incarcerated men in Brazil. Every 4 months, participants were assessed for symptoms and incident TB (Xpert or culture positive). Compared with remaining asymptomatic, developing symptoms was associated with higher TB risk (aRR 2.31; 95% CI 1.55–3.43), strongest among those with prior radiographic abnormalities (aRR 2.49; 95% CI 1.56–3.97), while remaining symptomatic was not associated (aRR 1.27; 95% CI .72–2.24). Longitudinal symptom monitoring may enhance TB screening in high-risk settings, particularly in complement with radiography.Keywords: tuberculosis, asymptomatic, prisons, active case finding, screening Tuberculosis (TB) continues to pose a major global health challenge [1]. Early detection and treatment are essential to interrupt transmission and improve outcomes, particularly in high-risk populations. Efforts to identify individuals living with TB often leverage symptom screening to prioritize individuals for confirmatory molecular rapid diagnostic tests [2]. However, conventional approaches to TB screening typically only assess symptoms at the time of screening entry, which may not capture the dynamic nature of symptom changes during TB disease progression [3, 4]. Symptoms that are typically classified as features of TB [5], such as cough, fever, night sweats, and weight loss, can fluctuate over the course of disease [6]. In high-risk populations such as persons deprived of liberty, these symptoms may also be caused by other common conditions, including upper respiratory tract infections, pneumonia, chronic obstructive pulmonary disease, and tobacco smoking [7]. Moreover, recent studies reveal that over 50% of individuals living with undiagnosed TB do not report symptoms at the time of diagnosis [1, 8]. Together, the nonspecific presentation of TB and high prevalence of asymptomatic TB may limit the ability of cross-sectional symptom screening to reliably identify individuals at imminent risk of active TB, resulting in continued transmission [8, 9]. Longitudinal monitoring of symptoms offers an opportunity to overcome some of these challenges by capturing individuals’ changes in symptom status over time, including new onset, persistence, or resolution. This may improve identification of individuals at high risk of TB, thereby facilitating the prioritization of confirmatory diagnostic testing. Prisons harbor some of the highest TB incidence and transmission rates worldwide [10, 11], and the World Health Organization recommends systematic screening be performed among incarcerated populations at least yearly [2]. However, active case finding is not widely implemented in most prisons in high-burden countries, partly due to resource constraints. Understanding how TB symptoms evolve in such settings could inform more resource-efficient screening approaches, yet longitudinal symptom trajectories in high-risk populations remain uncharacterized. To address this knowledge gap, we characterized longitudinal symptom trajectories in a cohort of incarcerated individuals who were evaluated for TB symptoms and tested for TB every four months. We assessed the relationship between symptom transitions and risk of incident TB. METHODS We performed systematic screening for TB every 4 months from February 2023 to September 2025 in 2 all-male prisons in the Brazilian state of Mato Grosso do Sul. After providing informed consent to participate, participants completed a standardized symptom questionnaire, administered by study nurses, to ascertain the presence of cough, fever, night sweats, weight loss, chest pain, or dyspnea. Presence of each symptom at each visit was assessed through yes/no questions (ie, “Do you have a cough?”). An individual was considered symptomatic if they confirmed the presence of any of these symptoms. At baseline, defined as the first visit in which a participant was enrolled in the study, all consenting participants had a chest radiograph performed and 2 sputum samples collected for testing by culture and Xpert® MTB/RIF Ultra (hereinafter “Xpert”). Radiographs were evaluated using automated detection software LunitTB (version v3.1.5.1), which outputs a numerical TB risk score between 0 and 100. At each 4-month follow-up visit, procedures were repeated: symptom assessment, a single sputum sample collection, and chest radiography. Symptom status transitions over sequential 4-month intervals were categorized as asymptomatic-to-symptomatic, symptomatic-to-asymptomatic, persistently symptomatic, or persistently asymptomatic. Incident TB was defined by at least 1 positive sputum culture or Xpert result. Trace Xpert results were repeated with a new sample. Those that persisted as traces were considered negative for TB and all trace samples were cultured. Eligibility was restricted to those with negative baseline Xpert and culture results who did not initiate TB treatment, regardless of radiographic findings, to ensure participants did not have prevalent TB. Less than 15 microbiologically negative participants initiated TB treatment based on clinical judgement. We only included participants who completed at least 1 follow-up visit to permit longitudinal symptom and incident TB assessment. Given the frequent transfers and movement within the prison system, nearly 25% of enrolled individuals missed follow-up visits intermittently. We excluded these individuals as gaps precluded reliable classification of symptom transitions across sequential 4-month intervals. We used the χ2 and Kruskal–Wallis tests to compare baseline demographic and clinical characteristics across symptom transition groups in the first 4-month interval. In supplementary analyses, we evaluated all 4-month symptom transitions across the study period to identify predictors associated with developing symptoms (if asymptomatic at the prior visit) or resolving symptoms (if symptomatic at the prior visit). We performed modified Poisson regression with robust standard errors clustered at the individual level to estimate adjusted risk ratios (aRRs) [12]. Models adjusted for the following baseline covariates: age, prison, prior TB, exposure to symptomatic cellmates, TB contact, current smoking status, any drug use, duration of incarceration, and binary LunitTB X-ray score using a threshold of 50. Specific symptom-level transition patterns were summarized descriptively by calculating the proportion of each individual symptom reported at visits where a transition occurred, distinguishing between new onset, resolution, and persistent symptoms. We also characterized how long participants remained in their initial symptom state during follow-up using survival analysis. For symptomatic individuals, the event was the first resolution of symptoms, while for asymptomatic individuals, the event was the first acquisition of symptoms. Participants were censored at TB diagnosis, loss to follow-up, or last study visit. For the primary analysis, we assessed whether symptom transitions during the preceding 4-month interval were associated with incident TB diagnosed at the subsequent visit, using the same modified Poisson framework described above but adjusting for radiographic status (Lunit score <50 vs ≥50) at the start of each interval rather than at baseline. Results were also stratified by radiographic status. Persistently asymptomatic individuals served as the reference category. All analyses were conducted in R (version 4.4.1). RESULTS After applying inclusion criteria, our study identified 2282 participants without TB at baseline who were assessed for symptoms at 2 or more consecutive visits between February 2023 and September 2025. The median baseline age of participants was 33 years (IQR 28–40), over 80% of participants were Black or mixed race (1854/2282), and most participants did not complete high school (83%) (Table 1). Table 1.Characteristics of Participants by Symptom Transition Between Baseline and First Follow-up Visit (N = 2282)CharacteristicsOverall(N = 2282)First 4-Month Interval Symptom Transition P ValueaPersistently Asymptomatic(n = 1417)Persistently Symptomatic(n = 215)Asymptomatic to Symptomatic(n = 330)Symptomatic to Asymptomatic(n = 320)Median age (IQR)33 (28, 40)33 (28, 40)33 (28, 40)32 (27, 39)33 (27, 40).208Incident TB50 (2.2)26 (1.8)4 (1.9)15 (4.5)5 (1.6).018Ethnicity.204 White419 (18.4)251 (17.7)40 (18.6)57 (17.3)71 (22.2) Black217 (9.5)127 (8.9)27 (12.6)33 (10.0)30 (9.4) Mixed1637 (71.7)1034 (73.0)145 (67.4)239 (72.4)219 (68.4) Asian8 (0.4)4 (0.3)3 (1.4)1 (0.3)0 (0) Indigenous1 (<0.1)1 (<0.1)0 (0)0 (0)0 (0)Education.562 Less than high school1893 (83.0)1175 (82.9)181 (84.2)279 (84.5)258 (80.6) Completed high school389 (17.0)242 (17.1)34 (15.8)51 (15.5)62 (19.4)Prison.256 EPJFC1745 (76.5)1089 (76.9)165 (76.7)239 (72.4)252 (78.8) IPCG537 (23.5)328 (23.1)50 (23.3)91 (27.8)68 (21.3)Previous TB409 (17.9)225 (15.9)54 (25.1)67 (20.3)63 (19.7).004TB contact1335 (58.5)818 (57.7)135 (62.8)204 (61.8)178 (55.6).239Symptomatic cellmates854 (37.4)415 (29.3)132 (61.4)126 (38.2)181 (56.6)<.001Current smoking1342 (58.8)741 (52.3)165 (76.7)220 (66.7)216 (67.5)<.001Any drug use1416 (62.1)829 (58.5)150 (69.8)226 (68.5)211 (65.9)<.001LunitTB Score at Baseline.961 <501584 (69.4)984 (69.4)146 (67.9)230 (69.7)224 (70.0) ≥50698 (30.6)433 (30.6)69 (32.1)100 (30.3)96 (30.0)Months incarcerated, median (IQR)12 (2, 44)13 (2, 41)14 (2, 71)11 (2, 34)9 (1, 61).213No (% of column n) unless otherwise specified. aPearson's χ2 test; Kruskal–Wallis test. Characteristics of Participants by Symptom Transition Between Baseline and First Follow-up Visit (N = 2282) No (% of column n) unless otherwise specified. aPearson's χ2 test; Kruskal–Wallis test. The majority of individuals reported no TB symptoms at baseline (1747/2,282, 76.5%) and 62.1% (1417/2282) remained asymptomatic at the first follow-up. Notably, 52% (26/50) of incident TB cases occurred among participants who remained asymptomatic. Among those asymptomatic at baseline, 14.5% developed symptoms by the first follow-up. Of the 23.4% symptomatic at baseline, 40% remained symptomatic and 60% reported resolution (Table 1). Individuals with persistent symptoms had the highest prevalence previous TB, exposure to symptomatic cellmates, current smoking, and any drug use (Table 1). Of 2282 total participants, 1725 (75.6%) were assessed for at least 3 visits, 1105 (48.4%) for at least 5, and 578 (25.3%) completed all 7 possible assessments. Among participants symptomatic at baseline (n = 535), the probability of remaining symptomatic declined steeply to 40.2% at the first follow-up and 7.6% by the last visit. In contrast, symptom absence was more stable: among those asymptomatic at baseline (n = 1749), 81.1% remained asymptomatic at the first follow-up and 43.8% were still asymptomatic at the last visit (Supplementary Figure 1). Symptom-level transitions revealed that cough (with or without sputum production) was the most frequently reported symptom among those developing symptoms (66%) and among those resolving symptoms (68%). Other common symptoms associated with symptom changes included chest pain, breathing difficulty, weight loss, and fever (Supplementary Figure 2). When evaluating predictors for symptom transitions among previously asymptomatic individuals, the onset of any TB symptom within the subsequent 4-month interval was associated with a previous TB diagnosis (aRR 1.26; 95% CI 1.08–1.47), current smoking (aRR 1.81; 95% CI 1.53–2.15), past smoking (aRR 1.34; 95% CI 1.09–1.66), any drug use (aRR 1.23; 95% CI 1.08–1.41), and exposure to symptomatic cellmates (aRR 1.30; 95% CI 1.15–1.47). Among previously symptomatic individuals, resolution of symptoms was less likely among those who reported current smoking (aRR 0.82; 95% CI .72–.92) or any drug use (aRR 0.88; 95% CI .79–.97) (Supplementary Table 1). In interaction models, these predictors did not significantly modify the association between symptom transitions and TB risk. Longitudinal analyses showed that compared with persistently asymptomatic individuals, transitioning from an asymptomatic to a symptomatic state within a 4-month interval was significantly associated with incident TB (aRR 2.31; 95% CI 1.55–3.43). In contrast, remaining symptomatic (aRR 1.27; 95% CI .72–2.24) or resolution from symptomatic to asymptomatic (aRR 1.13; 95% CI .68–1.90) were not significantly associated with TB risk, compared with those who stayed persistently asymptomatic (Figure 1). Stratified analyses revealed that the elevated risk among those developing symptoms was primarily driven by individuals with abnormal chest radiographs (aRR 2.49; 95% CI 1.56–3.97), though a similar trend was observed among those with normal X-rays (aRR 1.97; 95% CI .94–4.14). These findings indicate that symptom emergence was associated with elevated TB risk, with the strongest effects observed among those with radiographic abnormalities (Figure 1). Adjusted risk of incident TB by symptom transitions over 4-month intervals, overall and stratified by radiographic status. N represents the total number of 4-month symptom transition observations across all follow-up intervals during the study period. Adjusted risk ratios (aRRs) represent adjusted risk of TB diagnosed by sputum culture or Xpert within the same 4-month interval as the symptom transition. Models adjust for baseline characteristics including age, prison, prior TB, smoking, drug use, duration of incarceration, symptomatic cellmates, and TB contact. The overall model adjusts for Lunit X-ray score category at the first visit of each interval. The stratified models adjust for the baseline characteristics and are separated by Lunit X-ray score category at the first visit of each interval. DISCUSSION In our longitudinal cohort study in 2 high-incidence prisons, we found that TB-related symptoms followed dynamic trajectories over time. Between enrollment and the first follow-up visit just 4 months later, nearly 15% developed new symptoms, and 60% of those symptomatic at baseline reported resolution, underscoring the fluctuating nature of symptom reporting in this population. Importantly, more than half of all incident TB cases (52%, 26/50) occurred among participants who remained asymptomatic during the first interval, highlighting the substantial limitations of symptom-based screening in this setting. Prior studies in prisons and other high-risk congregate settings have similarly documented that reliance on cross-sectional symptom screening alone misses a considerable fraction of active TB cases, supporting the need for repeated active case finding [4, 13]. Our intensive screening program combining sputum testing and radiography every 4 months successfully detected microbiologically confirmed asymptomatic cases. Specifically, our analyses demonstrate that participants who developed new symptoms had more than double the risk of TB diagnosis compared with those who remained asymptomatic, while persistent symptoms conferred no excess risk. This pattern suggests that symptom emergence, rather than chronic symptom presence, offers greater resolution for identifying short-term risk. The lack of association with persistent symptoms may reflect alternative etiologies, such as concomitant respiratory infections, which could obscure the relationship with incident TB. Stratified analyses show that symptom emergence most strongly predicts TB risk among those with radiographic abnormalities. The overall association with symptom emergence has practical implications for resource-limited settings where radiography may be unavailable, while settings with radiography could combine repeated symptom assessments with X-rays for more precise targeting of diagnostic resources. Our study had several limitations. Symptom reporting relied entirely on self-report, which may be influenced by recall bias and subjective differences in symptom perception. Individuals who typically feel well may be more attuned to new symptoms, while those with chronic symptoms may have reduced sensitivity to incremental changes. However, our longitudinal assessment of symptom changes uses each individual as their own reference, potentially reducing the contextual and interviewer variability inherent in cross-sectional symptom reporting [14]. Additionally, the 4-month interval between screenings may not have captured shorter-term or transient symptom fluctuations. Excluding participants with intermittently missing visits may have introduced bias, as they were more likely to report drug use, lower education, and prison transfer or refusal. We also lacked reliable data on nutrition and longitudinal smoking behavior, precluding assessment of these important TB risk factors. Despite these limitations, our findings offer important insights into the temporal dynamics of TB symptoms and their potential utility in prioritizing individuals at highest risk for diagnostic screening and treatment. In conclusion, longitudinal assessment of TB-related symptoms reveals dynamic patterns of emergence and resolution that identify individuals at elevated short-term risk of active disease. Algorithms prioritizing individuals with either new-onset symptoms or radiographic abnormalities for molecular testing may improve sensitivity and cost-effectiveness in high-transmission, resource-limited settings. Recent evidence suggests that symptom screening, while less sensitive than annual radiography, can achieve comparable impact when applied frequently and at lower cost [15]. Monitoring symptom transitions may enhance this existing approach. Future studies should assess how integrating longitudinal symptom monitoring, radiography, and rapid diagnostics can be optimized to reduce transmission across high-burden settings. Supplementary Material Notes Acknowledgments. E. J., J. R. A., and J. C. conceived of the study. J. R. A., J. C., and R. D. O. developed the study plans. A. M. d. S. and A. d. S. S. enrolled participants and collected data. D. H. T. performed data management. E. J., J. V. B. B., Y. L., A. D. S., and J. R. A. designed and performed the data analysis. E. J. and J. V. B. B. wrote the manuscript. All authors contributed to revising the manuscript. The authors acknowledge the use of ChatGPT (GPT-4, OpenAI) for troubleshooting coding errors and minor grammar checking. AI was not used for generation of conclusions or interpretation of results. All content and analyses were independently verified by the authors, who take full responsibility for the manuscript. Data availability. Analytic code is available at https://github.com/Andrews-Lab-Stanford/LongitudinalTBSymptoms_BrazilPrisons.git. Financial support. This work was supported by the Ministério da Saúde of Brazil; Conselho Nacional de Desenvolvimento Científico e Tecnológico [CNPq; 404182/2019-4 to J. C.] and the U.S. National Institutes of Health [R01AI130058, K24AI182647 to J. R. A.]. Ethics approval. Ethical approval was obtained from the Research Ethics Committee of the Federal University of Mato Grosso do Sul, the Brazilian National Commission for Ethics in Research, and the Stanford University Research Ethics Committee. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed. Contributor Information Esther Jung, Department of Medicine, Stanford University School of Medicine, Stanford, California, USA. José Victor Bortolotto Bampi, Faculdade de Medicina, Federal University of Mato Grosso do Sul, Campo Grande, Brazil. Alessandra Moura da Silva, Faculdade de Medicina, Federal University of Mato Grosso do Sul, Campo Grande, Brazil. Yiran E Liu, Department of Medicine, Stanford University School of Medicine, Stanford, California, USA; SEICHE Center for Health and Justice, Yale University School of Medicine, New Haven, Connecticut, USA; Justice Collaboratory, Yale Law School, New Haven, Connecticut, USA. Daniel Henrique Tsuha, Faculdade de Medicina, Federal University of Mato Grosso do Sul, Campo Grande, Brazil. Argita D Salindri, Department of Medicine, Stanford University School of Medicine, Stanford, California, USA. Andrea da Silva Santos, Department of Health Sciences, Federal University of Grande Dourados, Dourados, Brazil. Roberto Dias de Oliveira, Department of Health Sciences, Federal University of Grande Dourados, Dourados, Brazil; Department of Nursing, State University of Mato Grosso do Sul, Dourados, Brazil. Julio Croda, Faculdade de Medicina, Federal University of Mato Grosso do Sul, Campo Grande, Brazil; Oswaldo Cruz Foundation, Campo Grande, Brazil; Department of Epidemiology of Microbial Diseases, Yale University School of Public Health, New Haven, Connecticut, USA. Jason R Andrews, Department of Medicine, Stanford University School of Medicine, Stanford, California, USA. Supplementary Data Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author. References Associated Data Supplementary Materials