Post-release tuberculosis risk among formerly incarcerated populations in Lima Peru

Abstract

People who experience incarceration face a markedly elevated risk of tuberculosis (TB), yet the extent to which this vulnerability persists after release remains incompletely characterized. We analyze data from a prospective cohort study conducted in Lima, Peru, including 3,658 adults with newly diagnosed pulmonary TB and 6,335 household contacts aged 15–60. Recent incarceration (within the preceding five years) is reported by 187 (5%) index patients and 112 household contacts (1.7%). Compared with those without incarceration history, formerly incarcerated individuals are more likely to present with more severe TB disease at diagnosis and experience a higher likelihood of adverse treatment outcomes. Among individuals with available incarceration timing data, TB diagnoses cluster after release, with nearly three-quarters occurring within two years. At the household level, TB infection prevalence at enrollment is higher among contacts with a history of incarceration and among those exposed to an index patient with prior incarceration, including among those with relatively short incarceration histories. These associations persist after adjustment for demographic, clinical, and socioeconomic factors. Together, these findings indicate that TB-related vulnerability extends beyond incarceration and into the post-release period, underscoring the importance of timely diagnosis and continuity of care post release.

Recent incarceration in Lima, Peru, was linked to more severe tuberculosis, worse TB treatment outcomes, and higher infection risk among household contacts. TB diagnoses clustered within two years of release, even after brief incarceration.

Introduction

Prisons are well recognized as high-risk environments for tuberculosis (TB) transmission. Structural conditions—including overcrowding, poor ventilation, prolonged close contact, and limited access to timely diagnosis and effective treatment—facilitate the spread of Mycobacterium tuberculosis (Mtb) among incarcerated populations. High prevalences of Human Immunodeficiency Virus (HIV) infection, malnutrition, substance use, and other comorbidities further increase susceptibility to TB disease. Consequently, TB incidence in prisons far exceeds that of surrounding communities, often by an order of magnitude or more^1–5.

While the elevated TB burden within prisons is well documented, the contribution of incarceration to TB burden beyond prison walls is less well characterized. High rates of prison entry and release, together with the long and variable latency between Mtb infection and active disease, mean that infection acquired during incarceration may not manifest until after release. Consistent with this possibility, epidemiologic studies from Latin America, Europe, and other settings report a marked increase in TB diagnoses among formerly incarcerated individuals shortly after release, with risk declining over time but remaining elevated for years^5–7. These observations suggest that incarceration-associated TB risk may persist well beyond the period of confinement.

In Peru, TB remains a major public health challenge, with one of the highest national incidence rates in the Americas and a persistent burden of drug-resistant disease⁸. Prisons occupy a central position within this epidemic. The Peruvian carceral system includes a substantial proportion of individuals held in pretrial detention⁹, resulting in a mixed population of pre-sentenced and sentenced individuals and wide variability in length of stay, with many pre-sentenced individuals detained for months to years¹⁰. The system operates at more than two to three times its intended capacity, reflecting chronic overcrowding and rapid population turnover⁹. Incarcerated individuals in Peru also experience a high prevalence of structural and clinical vulnerabilities, including malnutrition and comorbid conditions such as diabetes mellitus, both of which are established risk factors for TB disease and progression^11,12. Together, these structural and health-related features shape patterns of exposure and movement between prisons and surrounding communities. In response to sustained transmission in carceral settings, Peru has implemented nationwide active TB screening in prisons using chest radiography and rapid molecular diagnostics, revealing a substantial burden of previously undiagnosed and asymptomatic disease¹³. Collectively, these features position prisons as a critical interface between institutional and community TB transmission in Peru.

Genomic and modeling studies further support this prison–community linkage. Whole-genome sequencing analyses from Brazil, Thailand, and Georgia show that individuals with a history of incarceration are overrepresented in large Mtb transmission clusters spanning both incarcerated and non-incarcerated populations, consistent with spillover from prisons into surrounding communities^14–17. In parallel, recent modeling work has estimated that a substantial fraction of community TB cases in high-burden settings may be attributable, directly or indirectly, to transmission associated with incarceration, even when incarceration prevalence in the general population is relatively low^1,6,18. Together, these findings indicate that prisons may act as amplifiers of TB epidemics, with effects that extend into the broader community.

Despite this growing body of evidence, important gaps remain. Most prior studies emphasize post-release TB incidence or transmission inferred from genomic or modeling approaches, leaving the clinical and household-level consequences of incarceration, key pathways linking prisons to community spillover, poorly characterized. It remains unclear whether a history of incarceration is associated with more advanced disease or poorer treatment outcomes after release, or whether incarceration-related TB risk extends to household members, who represent a key interface between prisons and the community. Finally, although incarceration duration is often assumed to modify TB risk, it is unclear whether even brief periods of incarceration are sufficient to generate sustained post-release risk^19,20.

To address these gaps, we leveraged a large prospective cohort of individuals with newly diagnosed pulmonary TB and their household contacts (HHCs) in Lima, Peru. Using detailed epidemiologic, clinical, and incarceration history data, we examined whether a recent history of incarceration was associated with (1) TB severity at diagnosis, (2) TB treatment outcomes, and (3) the prevalence of TB infection among HHCs. By integrating analyses of clinical outcomes, timing of post-release diagnosis, and household-level infection patterns, this study aims to clarify how incarceration-related TB risk manifests after release and to better define the role of formerly incarcerated individuals in shaping community TB dynamics.

Results

Study population and definition of incarceration history

Between September 2009 and September 2012, we conducted a cohort study in Lima, Peru. We first enrolled 4044 newly diagnosed, microbiologically confirmed pulmonary TB index patients and their 12,767 HHCs.

Incarceration history among index patients

We restricted analyses to 3658 index patients aged 15–60 years, as no patients in other age groups reported a history of incarceration during the 60 months preceding enrollment (Fig. 1). Of these, 187 (5%) reported incarceration. Compared with those without incarceration history, index patients with prior incarceration were more likely to be male, report tobacco and alcohol use, have lower socioeconomic status, and present with a positive sputum smear at diagnosis (Table 1). Drug resistance profiles did not differ by incarceration history (Table 1). Among the 187 index patients who reported incarceration, 172 (92%) provided at least the year-level of prison entry and release information, whereas 138 (74%) provided complete month- and year-level dates (Fig. 1). Patients with complete incarceration date information did not differ significantly from those with incomplete (n = 34) or missing date (n = 15) across measured characteristics (Table S1 and Fig. 1).

Fig. 1. Study flow diagram for the analysis of post-release tuberculosis risk.

Flow diagram showing the derivation of the analytic sample of formerly incarcerated index patients used to evaluate the timing of pulmonary tuberculosis (TB) diagnosis following release from incarceration. Of 4044 microbiologically confirmed TB index patients enrolled in the parent prospective cohort study, 378 were excluded for age >60 years and 8 for missing incarceration history, leaving 3666 patients aged ≤60 years. Among these, 3471 reported no history of incarceration, and 187 reported a prior incarceration. Of the 187 formerly incarcerated patients, 15 did not provide any information on the timing of incarceration and were excluded; of the remaining 172 with duration information, 34 reported only year-level release dates and 138 reported complete month- and year-level entry and release dates. The 138 patients with complete date information constitute the analytic sample for Figs. 2 and 3.

Table 1.

Characteristics of index tuberculosis (TB) patients aged 15–60, stratified by recent incarceration history

Total			Without an incarceration history		With an incarceration history		P values
Age (n = 3658)							<0.001
16–30	2256	61.7%	2145	61.8%	111	59.4%
30–45	884	24.2%	814	23.5%	70	37.4%
>45	518	14.2%	512	14.8%	6	3.2%
Sex (n = 3658)							<0.001
Female	1396	38.2%	1392	40.1%	4	2.1%
Male	2262	61.8%	2079	59.9%	183	97.9%
Smoking status (n = 3583)							<0.001
Non-smoker	3485	97.3%	3328	97.7%	157	88.7%
Smoker	98	2.7%	78	2.3%	20	11.3%
Drinking status (n = 3495)							<0.001
Non-drinker	1913	54.7%	1877	56.5%	36	20.7%
Drinker	1582	45.3%	1444	43.5%	138	79.3%
Cough duration before diagnosis (n = 3595)							0.479
<2 weeks	861	23.9%	825	24.2%	36	19.5%
2–4 weeks	1204	33.5%	1139	33.4%	65	35.1%
4–8 weeks	817	22.7%	770	22.6%	47	25.4%
>8 weeks	713	19.8%	676	19.8%	37	20.0%
Sputum smear status (n = 3658)							0.053
Negative	1011	27.6%	971	28.0%	40	21.4%
Positive	2647	72.4%	2500	72.0%	147	78.6%
Cavitary disease (n = 3590)							0.355
No	2605	72.6%	2475	72.7%	130	69.5%
Yes	985	27.4%	928	27.3%	57	30.5%
Drug-resistant profiles (n = 3262)							0.319
Sensitive	1952	59.8%	1843	59.7%	109	61.9%
Resistance	830	25.4%	793	25.7%	37	21.0%
Multidrug-resistance	447	14.7%	418	14.5%	29	17.0%
Socioeconomic status (n = 3541)							0.039
Low	1226	34.6%	1155	34.4%	71	39.4%
Medium	1170	33.0%	1104	32.8%	66	36.7%
High	1145	32.3%	1102	32.8%	43	23.9%
TB history (n = 3647)							<0.001
No	2968	81.4%	2868	82.9%	100	53.5%
Yes	679	18.6%	592	17.1%	87	46.5%
Diabetes (n = 3621)							>0.999
No	3473	95.9%	3296	95.9%	177	96.2%
Yes	148	4.1%	141	4.1%	7	3.8%
Human immunodeficiency virus status (n = 3609)							0.226
No	3475	96.3%	3300	96.4%	175	94.6%
Yes	134	3.7%	124	3.6%	10	5.4%
Treatment outcomes (n = 3394)							<0.001
Cure	2884	85.0%	2775	86.1%	109	63.7%
Death	67	2.0%	62	1.9%	5	2.9%
Loss to follow-up	392	11.5%	340	10.5%	52	30.4%
Treatment failure	51	1.5%	46	1.4%	5	2.9%

P values were obtained from two-sided chi-square tests; continuous variables were categorized prior to analysis. P values are reported without adjustment for multiple comparisons.

Incarceration history and TB severity at diagnosis

We evaluated whether the index patients who had been incarcerated had higher TB disease severity upon diagnosis. We assessed TB severity at enrollment using a modified Bandim TB Severity Score²¹, incorporating respiratory symptoms, fever, and body mass index (BMI), with higher scores indicating more severe disease. The seven-item TB severity score had a mean of 2.75 (standard deviation [SD], 1.39; range, 0–7) and showed low internal consistency (Cronbach’s α = 0.47; 95% CI: 0.44–0.50), with a low average inter-item correlation (r̄ = 0.11), reflecting the inclusion of clinically distinct disease manifestations. Removal of any single item did not materially improve reliability (α range: 0.37–0.48). These results indicate that the score captures heterogeneous dimensions of TB severity and is best interpreted as a composite clinical index rather than a unidimensional psychometric scale.

In univariable analyses, incarceration history was associated with higher TB severity scores at diagnosis (mean score difference [MSD] = 0.29; 95% CI: 0.08– 0.51), corresponding to a modest standardized mean difference (SMD = 0.21). This association persisted after adjustment for potential confounders, including age, sex, socioeconomic status, diabetes mellitus, and HIV infection (MSD = 0.26; 95% CI: 0.04–0.48). After further adjustment for cough duration prior to diagnosis, a marker of diagnostic delay that may act as a mediator in the observed association, the effect estimate was attenuated and no longer statistically significant (MSD = 0.17; 95% CI: −0.04 to 0.37) (Table 2). In sensitivity analyses excluding BMI, the magnitude and direction of associations were similar, and the standardized mean difference remained nearly unchanged (SMD = 0.21) (Table 2). Together, these results indicate that a history of incarceration is associated with more severe TB at presentation, an association that is partly explained by delays in diagnosis.

Table 2.

Univariable and multivariable linear regression models examining associations between the incarceration history of index patients and their tuberculosis (TB) disease severity score at diagnosis

		Univariable (N = 3658)		Multivariable 1 (N = 3476)		Multivariable 2 (N = 3421)
		Mean score difference (95% CI)	P values	Mean score difference (95% CI)	P values	Mean score difference (95% CI)	P values
Incarceration history
	No	Ref		Ref		Ref
	Yes	0.29 (0.08 to 0.51)	0.006	0.26 (0.04 to 0.48)	0.020	0.17 (−0.04 to 0.37)	0.112
Age
	16–30	Ref		Ref		Ref
	30–45	−0.13 (−0.24 to −0.03)	0.015	−0.16 (−0.27 to −0.05)	0.006	−0.21 (−0.31 to −0.11)	<0.001
	>45	−0.15 (−0.27 to −0.02)	0.027	−0.12 (−0.26 to 0.02)	0.102	−0.2 (−0.34 to −0.06)	0.004
Sex
	Female	Ref		Ref		Ref
	Male	0.2 (0.11 to 0.29)	<0.001	0.17 (0.07 to 0.26)	0.001	0.17 (0.08 to 0.27)	<0.001
Socioeconomic status
	Low	Ref		Ref		Ref
	Medium	−0.14 (−0.25 to −0.03)	0.013	−0.11 (−0.22 to 0)	0.046	−0.08 (−0.18 to 0.03)	0.138
	High	−0.25 (−0.36 to −0.13)	<0.001	−0.22 (−0.33 to −0.11)	<0.001	−0.17 (−0.28 to −0.07)	0.001
Diabetes
	No	Ref		Ref		Ref
	Yes	−0.2 (−0.42 to 0.01)	0.063	−0.1 (−0.34 to 0.14)	0.405	−0.05 (−0.27 to 0.18)	0.690
Human immunodeficiency virus status
	No	Ref		Ref		Ref
	Yes	0.52 (0.27 to 0.77)	<0.001	0.54 (0.29 to 0.79)	<0.001	0.53 (0.3 to 0.77)	<0.001
Cough duration before diagnosis
	<2 weeks	Ref				Ref
	2–4 weeks	1 (0.88 to 1.12)	<0.001			0.99 (0.86 to 1.11)	<0.001
	4–8 weeks	1.18 (1.05 to 1.31)	<0.001			1.15 (1.02 to 1.29)	<0.001
	>8 weeks	1.34 (1.21 to 1.48)	<0.001			1.36 (1.22 to 1.5)	<0.001

Effect estimates are presented as mean score differences with 95% confidence intervals. Multivariable model 1 adjusted for age, sex, socioeconomic status, diabetic status, and human immunodeficiency virus status. Multivariable model 2 additionally adjusted for cough duration before diagnosis. P values are from two-sided tests and are reported without adjustment for multiple comparisons.

Incarceration history and TB treatment outcomes

We next evaluated whether a history of incarceration was associated with TB treatment outcomes, defined according to World Health Organization criteria (death, loss to follow-up, or treatment failure). Secondary analyses focused on biologically adverse outcomes, defined as death or treatment failure.

Of 3658 index patients, 264 (7%) were excluded due to clinical default or withdrawal, leaving 3394 patients for analysis. Among these, 2884 (85%) experienced a favorable treatment outcome. Of the 510 patients with poor outcomes, 67 (13%) died, 392 (76%) were lost to follow-up, and 51 (10%) experienced treatment failure. Across all categories of unfavorable outcomes, patients with a history of incarceration had a higher proportion of poor outcomes than those without such a history. This absolute difference was most pronounced for loss to follow-up (30.4% vs. 11.5%), whereas differences were smaller for death (2.9% vs. 2.0%) and treatment failure (2.9% vs. 1.5%) (Table 1).

Index patients with a history of incarceration had a higher risk of WHO-defined poor treatment outcomes compared with those without such a history (univariable risk ratio [RR] = 2.61; 95% CI: 2.10–3.24). This association persisted after adjustment for potential confounders (adjusted RR = 1.69; 95% CI: 1.29–2.22) and after further adjustment for potential mediators (RR = 1.86; 95% CI: 1.41–2.47) (Table 3).

Table 3.

Univariable and multivariable modified Poisson regression models examining associations between the incarceration history of index patients and unfavorable tuberculosis (TB) treatment outcomes

		Univariable (N = 3402)		Multivariable 1 (N = 3041)		Multivariable 2 (N = 2668)*
		Risk ratio (95% CI)	P values	Risk ratio (95% CI)	P values	Risk ratio (95% CI)	P values
Incarceration history
	No	Ref		Ref		Ref
	Yes	2.61 (2.1–3.24)	<0.001	1.69 (1.29–2.22)	<0.001	1.86 (1.41–2.47)	<0.001
Age
	16–30	Ref		Ref		Ref
	30–45	1.08 (0.89–1.3)	0.438	0.81 (0.66–1)	0.046	0.81 (0.65–1.01)	0.057
	>45	1.07 (0.85–1.35)	0.572	0.94 (0.71–1.25)	0.678	0.89 (0.65–1.22)	0.47
Sex
	Female	Ref		Ref		Ref
	Male	1.82 (1.51–2.19)	<0.001	1.32 (1.08–1.61)	0.008	1.35 (1.09–1.67)	0.007
Smoking status
	Non-smoker	Ref		Ref		Ref
	Smoker	1.94 (1.38–2.73)	<0.001	1.24 (0.83–1.83)	0.291	1.12 (0.73–1.72)	0.592
Drinking status
	Non-drinker	Ref		Ref		Ref
	Drinker	1.76 (1.49–2.09)	<0.001	1.47 (1.23–1.77)	<0.001	1.48 (1.22–1.8)	<0.001
Cough duration before diagnosis
	less than 2 weeks	Ref				Ref
	2–4 weeks	0.94 (0.75–1.18)	0.582			0.89 (0.69–1.15)	0.363
	4–8 weeks	1.12 (0.89–1.42)	0.333			1.04 (0.79–1.35)	0.792
	>8 weeks	1.27 (1–1.61)	0.05			1.06 (0.81–1.4)	0.655
Sputum smear status
	Negative	Ref				Ref
	Positive	1.23 (1.02–1.49)	0.03			1.14 (0.92–1.42)	0.234
Cavitary disease
	No	Ref				Ref
	Yes	1.07 (0.9–1.28)	0.442			1.06 (0.87–1.28)	0.574
Drug-resistant profiles
	Sensitive	Ref				Ref
	Resistance	1.36 (1.07–1.73)	<0.001			1.39 (1.1–1.75)	0.006
	Multidrug-resistance	3.7 (3.02–4.52)	<0.001			3.3 (2.68–4.05)	<0.001
Socioeconomic status
	Low	Ref		Ref		Ref
	Medium	0.94 (0.78–1.14)	0.525	1.05 (0.86–1.28)	0.613	1.04 (0.84–1.28)	0.725
	High	0.77 (0.62–0.94)	0.011	0.87 (0.7–1.07)	0.187	0.94 (0.75–1.17)	0.567
TB history
	No	Ref		Ref		Ref
	Yes	2.22 (1.89–2.61)	<0.001	1.83 (1.51–2.21)	<0.001	1.48 (1.21–1.82)	<0.001
Diabetes
	No	Ref		Ref		Ref
	Yes	1.1 (0.75–1.63)	0.619	1.17 (0.74–1.87)	0.504	1.33 (0.81–2.18)	0.258
Human immunodeficiency virus status
	No	Ref		Ref		Ref
	Yes	2.48 (1.92–3.2)	<0.001	2.36 (1.8–3.1)	<0.001	2.16 (1.59–2.94)	<0.001

^*Further adjusted for cough duration before diagnosis, sputum smear status, cavitary disease, and drug-resistant profiles.

Effect estimates are risk ratios with 95% confidence intervals. Multivariable model 1 adjusted for demographic, behavioral, and comorbidity covariates; multivariable model 2 additionally adjusted for clinical and disease-related characteristics, as listed. P values are two-sided and unadjusted for multiple comparisons.

When analyses were restricted to biologically adverse outcomes (including death and treatment failure but excluding loss to follow-up), effect estimates increased in magnitude. In models adjusted for confounders, incarceration history remained associated with a higher risk of adverse outcomes (adjusted RR = 2.08; 95% CI: 1.05–4.12), with further strengthening after adjustment for potential mediators (RR = 2.93; 95% CI: 1.37–6.29) (Table S2).

Together, these findings indicate that incarceration history is associated with substantially worse TB treatment outcomes, particularly for biologically adverse endpoints. Notably, compared with univariable models, the estimated effects of both incarceration history and drug-resistance profile increased after multivariable adjustment, despite drug resistance being considered a potential mediator of this association. This pattern likely reflects the fact that incarceration history and drug resistance capture overlapping yet distinct biological and social processes; mutual adjustment may therefore remove shared variance, yielding stronger independent associations for each factor.

Timing of TB diagnosis following release from incarceration

To assess whether TB disease among previously incarcerated individuals was plausibly linked to infection or delayed diagnosis during incarceration, we examined the interval, measured in months, between prison release and TB diagnosis. Because we only assessed individuals who developed active TB disease, we examined the distribution of time from release to diagnosis among cases rather than fitting time-to-event models.

Of the 138 index patients who reported incarceration release dates, the median time from release to TB diagnosis was 11 months (95% CI: 7.5–15.2), substantially shorter than the 30-month median expected under a uniform distribution of diagnoses across the 60-month exposure window; 73% of TB diagnoses occurred within 24 months of release (Figs. 2 and 3), and a Kolmogorov–Smirnov test confirmed significant deviation from a uniform distribution (P < 0.001).

Fig. 2. Empirical cumulative distribution of time from incarceration release to tuberculosis diagnosis.

Empirical cumulative distribution function (ECDF) of time from incarceration release to pulmonary tuberculosis (TB) diagnosis among n = 138 formerly incarcerated index patients with complete month- and year-level incarceration entry and release dates. The black step function shows the observed cumulative distribution of times to TB diagnosis; the red dashed line indicates a uniform reference distribution over the 0–60-month exposure window. The x axis shows months since release (0–60 months), and the y axis shows the cumulative proportion of patients diagnosed. Deviation of the observed distribution from the uniform reference was assessed using a two-sided one-sample Kolmogorov–Smirnov test (P < 0.001). No adjustment for multiple comparisons was applied. Source data are provided as a Source Data file.

Fig. 3. Distribution of time from incarceration release to tuberculosis diagnosis.

Histogram of the proportion of post-release pulmonary tuberculosis (TB) diagnoses occurring within successive 6-month intervals after release from incarceration, among n = 138 formerly incarcerated index patients with complete month- and year-level incarceration entry and release dates. The x axis shows months since release in 6-month bins from 0 to 60 months; the y axis shows the percentage of TB diagnoses occurring within each interval, expressed as a proportion of the 138 patients. Deviation from a uniform distribution over the 60-month window was assessed using a two-sided one-sample Kolmogorov–Smirnov test (P < 0.001). Source data are provided as a Source Data file.

Given the high TB exposure burden in prison settings, we next evaluated whether relatively short periods of incarceration were sufficient to influence the timing of post-release diagnoses. Among the 72 (53%) individuals incarcerated for ≤12 months, the clustering of diagnoses shortly after release persisted (median time to diagnosis =  12 months; 95% CI: 11.5–21.0) (Fig. S1), indicating that the observed pattern was not driven solely by prolonged incarceration.

As an additional plausibility check, we further restricted the analysis to individuals with extremely short incarceration durations (≤1 month; n = 13). In this subgroup, the median time from release to TB diagnosis increased to 21 months, approaching the 30-month median expected under a uniform distribution. Because this analysis was limited by the small sample size, this attenuation toward the null was expected and is compatible with the hypothesis that incarceration duration may influence post-release TB risk, although estimates were imprecise.

This observed clustering of diagnoses shortly after release supports the interpretation that incarceration contributes to TB risk that persists into the post-release period.

Incarceration exposure and the prevalence of TB infection among household contacts

We next assessed whether the HHCs who had been incarcerated in the previous 5 years had a higher prevalence of TB infection. We restricted the analyses among 6335 HHCs and of these 112 (2%) reported an incarceration history, and 368 (6%) were exposed to an index patient with an incarceration history (Tables 4, S3, and S4 and Fig. S2). HHCs who had previously been incarcerated had a higher prevalence of TB infection at enrollment compared with those without such a history (prevalence risk ratio [PRR] = 1.29; 95% CI: 1.16–1.45) (Table 5). HHCs exposed to an index patient with an incarceration history had a higher prevalence of TB infection (PRR = 1.14; 95% CI: 1.02–1.28). These associations were almost unaltered after adjusting for potential confounders (adjusted PRR for HHC incarceration = 1.27; 95% CI: 1.11–1.46; adjusted PRR for index patient incarceration = 1.13; 95% CI: 1.00–1.28) (Table 5).

Table 4.

Characteristics of household contacts (HHCs) aged 15–60, stratified by tuberculosis (TB) infection status

		Total		Non-TB-infected		TB-infected		P values
HHC incarceration history (N = 6335)								<0.001
	No	6223	98%	2952	99%	3271	98%
	Yes	112	2%	27	1%	85	3%
Index incarceration history (N = 6335)								0.003
	No	5967	94%	2834	95%	3133	93%
	Yes	368	6%	145	5%	223	7%
HHC age (N = 6335)								<0.001
	16–30	3041	48%	1653	56%	1388	41%
	30–45	1864	29%	800	27%	1064	32%
	>45	1430	23%	526	18%	904	27%
HHC sex (N = 6335)								0.919
	Female	3729	59%	1756	59%	1973	59%
	Male	2606	41%	1223	41%	1383	41%
HHC number of Bacillus Calmette–Guérin scars (N = 6333)								<0.001
	0	568	9%	300	10%	268	8%
	1	3590	57%	1762	59%	1828	55%
	≥2	2175	34%	916	31%	1259	38%
HHC socioeconomic status (N = 6166)								0.176
	Low	2121	34%	966	33%	1155	35%
	Medium	2706	44%	1291	44%	1415	43%
	High	1339	22%	648	22%	691	21%
Index age (N = 6335)								0.053
	16–30	4375	69%	2099	71%	2276	68%
	30–45	1157	18%	511	17%	646	19%
	>45	803	13%	369	12%	434	13%
Index sex (N = 6335)								0.38
	Female	2458	39%	1173	39%	1285	38%
	Male	3877	61%	1806	61%	2071	62%
Index smoking status (N = 6236)								0.069
	Non-smoker	6040	97%	2847	97%	3193	97%
	Smoker	196	3%	79	3%	117	4%
Index drinking status (N = 6084)								0.071
	Non-drinker	3358	55%	1626	56%	1732	54%
	Drinker	2726	45%	1256	44%	1470	46%
Index human immunodeficiency virus status (N = 6292)								0.768
	No	6102	97%	2876	97%	3226	97%
	Yes	190	3%	87	3%	103	3%
Index diabetes (N = 6288)								0.192
	No	6043	96%	2852	96%	3191	96%
	Yes	245	4%	105	4%	140	4%
Index sputum smear status (N = 6335)								0.012
	Negative	1757	28%	871	29%	886	26%
	Positive	4578	72%	2108	71%	2470	74%
Index cough duration before diagnosis (N = 6235)								<0.001
	<2 weeks	1427	23%	747	25%	680	21%
	2–4 weeks	2120	34%	961	33%	1,159	35%
	4 weeks to 8 weeks	1392	22%	664	23%	728	22%
	>8 weeks	1296	21%	566	19%	730	22%
Index cavitary disease (N = 6251)								0.182
	No	4538	73%	2163	73%	2375	72%
	Yes	1713	27%	784	27%	929	28%

P values were obtained from two-sided chi-square tests. P values are reported without adjustment for multiple comparisons.

Table 5.

Univariable and multivariable modified Poisson regression models examining associations between incarceration history of household contacts (HHCs) or their index patients and prevalent tuberculosis (TB) infection at enrollment

		Univariable (N = 6335)		Multivariable (N = 5618)
		Prevalence risk ratio (95% CI)	P values	Prevalence risk ratio (95% CI)	P values
HHC incarceration history
	No	Ref		Ref
	Yes	1.29 (1.16–1.45)	<0.001	1.27 (1.11–1.46)	0.001
Index incarceration history
	No	Ref		Ref
	Yes	1.14 (1.02–1.28)	0.018	1.13 (1–1.28)	0.054
HHC age
	≤30	Ref		Ref
	30–45	1.26 (1.19–1.34)	<0.001	1.24 (1.17–1.32)	<0.001
	>45	1.37 (1.3–1.45)	<0.001	1.34 (1.26–1.42)	<0.001
HHC sex
	Female	Ref		Ref
	Male	0.98 (0.93–1.03)	0.368	0.99 (0.94–1.04)	0.761
HHC number of Bacillus Calmette–Guérin scars
	0	Ref		Ref
	1	1.08 (0.99–1.19)	0.073	1.09 (0.99–1.2)	0.079
	≥2	1.23 (1.12–1.35)	<0.001	1.19 (1.08–1.32)	0.001
HHC socioeconomic status
	Low	Ref		Ref
	Medium	0.94 (0.89–1.01)	0.074	0.96 (0.9–1.03)	0.23
	High	0.94 (0.87–1.01)	0.087	0.94 (0.87–1.01)	0.095
Index age
	≤30	Ref		Ref
	30–45	1.07 (1–1.15)	0.036	1.05 (0.98–1.13)	0.185
	>45	1.04 (0.96–1.13)	0.29	1.03 (0.94–1.14)	0.481
Index sex
	Female	Ref		Ref
	Male	1.01 (0.96–1.07)	0.673	0.97 (0.92–1.04)	0.421
Index smoking status
	Non-smoker	Ref		Ref
	Smoker	1.13 (0.97–1.31)	0.108	1.03 (0.87–1.21)	0.741
Index drinking status
	Non-drinker	Ref		Ref
	Drinker	1.05 (0.99–1.11)	0.087	1.03 (0.97–1.09)	0.397
Index human immunodeficiency virus status
	No	Ref		Ref
	Yes	1.02 (0.89–1.17)	0.784	1.06 (0.91-1.23)	0.477
Index diabetes
	No	Ref		Ref
	Yes	1.1 (0.96–1.25)	0.16	0.98 (0.83–1.15)	0.772
Index sputum smear status
	Negative	Ref		Ref
	Positive	1.06 (1–1.13)	0.047	1.03 (0.97–1.1)	0.358
Index cough duration before diagnosis (N = 6235)
	<2 weeks	Ref		Ref
	2–4 weeks	1.14 (1.06–1.24)	0.001	1.12 (1.03–1.21)	0.009
	4–8 weeks	1.11 (1.02–1.21)	0.018	1.08 (0.99–1.18)	0.085
	>8 weeks	1.18 (1.08–1.29)	<0.001	1.14 (1.04–1.25)	0.005
Index cavitary disease
	No	Ref		Ref
	Yes	1.06 (1–1.12)	0.066	1.05 (0.99–1.12)	0.108

Effect estimates are prevalence risk ratios with 95% confidence intervals, accounting for within-household clustering. The multivariable model adjusted for HHC and index demographic, clinical, and behavioral characteristics as listed. P values are two-sided and unadjusted for multiple comparisons.

Given the high burden of TB exposure in prison settings, we next evaluated whether even relatively short periods of incarceration were associated with increased TB infection risk among HHCs. We found that HHCs with short incarceration stays were associated with increased TB infection prevalence compared with no incarceration (≤6 months vs. none: aPRR = 1.25; 95% CI: 1.01–1.55; ≤12 months vs. none: aPRR = 1.28; 95% CI: 1.06–1.56) (Table S5).

These results indicate that incarceration-associated TB risk extends beyond the individual, increasing the likelihood of TB infection among HHCs even after relatively short periods of incarceration.

Discussion

In this prospective cohort study, index TB patients with a recent history of incarceration presented with higher TB severity scores at diagnosis and experienced worse treatment outcomes than those without such a history. Among previously incarcerated index patients with available timing data, TB diagnoses clustered in the period following release, with over 70% occurring within two years; this pattern was also evident among individuals incarcerated for less than one year. At the household level, a higher prevalence of TB infection at enrollment was apparent among HHCs with a recent incarceration history and among contacts exposed to an index patient with a history of incarceration. Similar patterns persisted among HHCs who reported relatively short periods of incarceration.

Our finding that a substantial proportion of TB patients with a recent incarceration history were diagnosed within two years after release is consistent with reports from diverse settings, including Brazil, Paraguay, and France. Mabud et al. found that TB incidence among formerly incarcerated individuals in the Brazilian state of Mato Grosso do Sul between 2007 and 2013 was approximately five times higher than baseline community incidence during the first year after release, with risk declining thereafter⁶. Sequera et al. reported that among individuals incarcerated in Paraguay between 2010 and 2021, more than 40% of TB cases were diagnosed after release; within this group, TB notifications were highest in the first year following release and remained elevated for several years⁵. Similarly, Niaux et al. observed that among individuals who developed TB after release in France between 2008 and 2020, over one-third were diagnosed within 2 years⁷. Together, these patterns are consistent with both undiagnosed disease at the time of release and progression following infection acquired during incarceration; neither prior studies nor our data permit distinction between these pathways.

To our knowledge, few studies have examined whether a history of incarceration is associated with TB severity at diagnosis or subsequent treatment outcomes. In our cohort, individuals with a recent incarceration history presented with more severe disease at diagnosis, a pattern that may reflect delays occurring both before and after release. Some individuals may have had undiagnosed TB during incarceration due to limited access to healthcare in prison settings¹¹, a possibility supported by active TB screening programs in Peruvian prisons that have identified substantial numbers of previously undiagnosed cases^13,22, while others may have developed disease after release but experienced delays in healthcare-seeking or access post release^23,24. In our analyses, adjustment for cough duration prior to diagnosis attenuated the association between incarceration history and TB severity, further supporting diagnostic delay as a contributing pathway. Such delays may also contribute to worse biologically adverse treatment outcomes, including death and treatment failure, through more advanced disease at treatment initiation. In our cohort, individuals with an incarceration history had higher rates of loss to follow-up, consistent with persistent social and structural barriers to sustained care after release^23,24. Notably, associations between incarceration history and adverse treatment outcomes strengthened after adjustment for drug-resistance profile, despite drug resistance being considered a potential mediator, suggesting that incarceration history and drug resistance reflect overlapping but distinct processes and that accounting for drug resistance clarifies the independent contribution of incarceration-related vulnerability to poor outcomes.

Despite strong evidence linking incarceration to elevated TB infection prevalence within prisons, few studies have examined TB infection prevalence among individuals after release. Prison-based screening studies consistently report substantially higher TB infection prevalence among incarcerated individuals than in the general population, although estimates vary widely across settings²⁵. Notably, TB infection prevalence among individuals with relatively short periods of incarceration has been reported to remain higher than background population estimates^19,26. Consistent with this pattern, we observed a higher prevalence of TB infection among HHCs with a history of incarceration, including those who reported relatively short incarceration durations. These findings highlight the potential for brief exposure to high-burden prison environments or associated social contexts to be associated with elevated TB infection prevalence, while underscoring the difficulty of attributing infection to specific settings or periods of exposure.

Taken together, our findings indicate that incarceration and reentry constitute a continuum of TB vulnerability rather than discrete periods of risk confined to the prison setting. The concentration of TB diagnoses after release, the association with more advanced disease at presentation, and the higher prevalence of TB infection among HHCs, including those linked to relatively short periods of incarceration, point to delays in diagnosis and disrupted engagement with care around the time of release. These patterns suggest that opportunities for timely TB detection may be missed both during incarceration and in the immediate post-release period, particularly in settings characterized by high population turnover and short or pre-trial detention. Although our data do not permit attribution of infection or disease progression to specific settings or periods of exposure, they identify the transition from incarceration to the community as a critical juncture at which diagnostic delays and gaps in linkage to care may contribute to adverse clinical and household-level outcomes. Future studies should evaluate whether strategies that strengthen TB detection and continuity of care around release can mitigate these vulnerabilities.

At a health-systems level, these findings underscore the challenges of managing TB across institutional boundaries. The observed patterns of more severe disease at diagnosis, poorer treatment outcomes, and elevated TB infection prevalence following release suggest persistent gaps in linkage, follow-up, and retention in care among justice-involved populations. While our study does not assess the effectiveness of specific interventions, it highlights the importance of coordination between correctional and community health services, particularly during the early post-release period. Addressing TB among justice-involved populations will likely require approaches that extend beyond prison-based activities to ensure continuity of care across settings.

Our study has several limitations. First, index TB patients were enrolled in a prospective cohort with structured follow-up, which may have contributed to better treatment outcomes than those observed in routine care settings. Because follow-up procedures were protocolized, differential follow-up by incarceration history is unlikely; however, unmeasured social factors associated with incarceration may have influenced engagement with care and contributed to observed differences in treatment outcomes. The former would be expected to attenuate associations, whereas the latter could introduce bias of uncertain direction. Second, all HHCs were exposed to TB within households, resulting in a higher baseline prevalence of TB infection among contacts without a history of incarceration than would be expected in the general population. Although we adjusted for index-patient risk factors for TB infection, residual confounding cannot be ruled out. Third, incarceration history was self-reported and subject to underreporting due to social desirability or recall error in reporting incarceration timing, which may have resulted in non-differential misclassification of exposure and further attenuation of observed associations. Fourth, our data do not allow attribution of TB infection or disease onset to specific settings or periods, and we were unable to distinguish disease present at the time of release from disease that developed after release. Also, timing and duration of incarceration were incompletely reported for a subset of participants, limiting temporal precision. Fifth, the generalizability of our findings may be limited to settings with epidemiologic and structural features similar to urban Peru. TB transmission dynamics, incarceration patterns, and prison conditions vary widely across countries, particularly between low- and middle-income and high-income settings; consequently, both the magnitude and mechanisms of post-release TB risk may differ in contexts with lower background incidence, longer incarceration durations, or more robust prison health systems. Finally, although we adjusted for a range of clinical and demographic factors, residual confounding by unmeasured social and structural determinants—particularly those affecting individuals during reentry—may persist.

In this prospective cohort study, recent incarceration was associated with more severe TB disease at diagnosis, poorer treatment outcomes, and a higher prevalence of TB infection among HHCs, including after relatively short periods of incarceration. These findings indicate that TB-related vulnerability extends beyond incarceration and into the transition back to the community, underscoring the importance of timely diagnosis and continuity of care during reentry.

Methods

Ethics statements

This study was approved by the Institutional Review Board of Harvard Medical School (Boston, MA, USA; protocol number 19332) and the Comité Institucional de Ética en Investigación del Instituto Nacional de Salud (CIEI-INS; Lima, Peru; protocol number 014-2008). All participants were community-dwelling adults and adolescents recruited at district health centers; the study did not involve enrollment of currently incarcerated individuals. Written informed consent was obtained from all adult participants (≥18 years) prior to enrollment. For participants aged 16–17 years, written informed assent was obtained from the participant in addition to written informed consent from a parent or legal guardian. For participants younger than 16 years, written informed assent was obtained from the participant together with written informed consent from a parent or legal guardian.

Study setting and data collection

Between 2009 and 2012, we conducted a prospective cohort study in 106 district health centers across Lima, Peru, which provide routine TB diagnosis and treatment under the national tuberculosis program. We systematically enrolled individuals aged ≥15 years with newly diagnosed pulmonary tuberculosis as index patients. We confirmed their disease status through sputum smear microscopy and/or mycobacterial culture tests. At enrollment, trained study staff collected standardized information on sociodemographic characteristics, prior TB, tobacco and alcohol use, incarceration history, duration of respiratory symptoms before diagnosis, and comorbidities, including HIV infection and diabetes mellitus. We measured height and weight and recorded TB-related symptoms, including cough, hemoptysis, night sweats, dyspnea, and fever. Chest radiographs were obtained at baseline and independently interpreted by two radiologists using a standardized form. Participants provided blood samples and consented to HIV testing with pre- and post-test counseling. Index patients received TB treatment according to national guidelines²⁷ and were followed longitudinally with repeat microbiologic testing and clinical assessments. At the end of follow-up, we abstracted detailed treatment histories and outcomes from health center medical charts.

Following enrollment of index patients, study teams visited households and enrolled consenting HHCs. At baseline, we collected data on age, sex, relationship to the index patient, housing characteristics, socioeconomic indicators, health behaviors, comorbidities, prior TB, and incarceration history. HHCs were screened for TB symptoms at enrollment; those reporting symptoms suggestive of active disease were referred to local health centers for clinical evaluation, including chest radiography and sputum testing. Among contacts without prior TB disease or a known positive tuberculin skin test (TST), TB infection status at enrollment was assessed using the TST. HHCs with a negative baseline TST and no evidence of TB disease underwent repeat testing at six and twelve months. Contacts who developed TB-related symptoms during follow-up were referred for evaluation for incident TB disease. We additionally reviewed health center records within study catchment areas to identify TB diagnoses among enrolled HHCs during follow-up.

Drug-susceptibility testing

Among 4044 microbiologically confirmed index TB patients, 3572 were culture positive. During the cohort study period, 3212 (90%) of these culture-positive isolates underwent drug-susceptibility testing (DST) to first-line anti-TB drugs, including isoniazid, rifampicin, ethambutol, streptomycin, and pyrazinamide. Isolates demonstrating resistance to any first-line drug subsequently underwent second-line DST using the agar plate method, including testing for kanamycin, capreomycin, ethionamide, ciprofloxacin, para-aminosalicylic acid, and cycloserine²⁸. Among the 360 culture-positive patients for whom DST was not performed during the cohort study period, drug-resistance profiles were retrospectively determined for 189 isolates using minimum inhibitory concentration (MIC) testing of archived sputum samples²⁹.

Incarceration history

At enrollment, index patients and HHCs reported any incarceration lasting more than one day during the 60 months preceding enrollment. Incarceration history was defined as any self-reported incarceration during this period, regardless of duration. For participants reporting incarceration, we recorded the start and end months and years of each episode. Using these data, we calculated the interval in months from prison release to TB diagnosis for index patients and the total duration of incarceration for HHCs in months.

TB severity score and TB treatment outcomes of index patients

We assessed TB severity in index patients using a modified Bandim TB Severity Score²¹, which incorporated the presence of cough, hemoptysis, night sweats, dyspnea, fever (>37 °C), and BMI < 18 kg/m². Each component contributed one point, with an additional point assigned for BMI < 16 kg/m², yielding a maximum score of seven. We followed index patients until they achieved TB cure, loss to follow-up, death, or a maximum of five years, whichever occurred first. We classified treatment outcomes in two ways. First, following World Health Organization definitions³⁰, outcomes were categorized as successful (cure or treatment completion) or unsuccessful (treatment failure, loss to follow-up, or death). Second, to focus on biologically meaningful outcomes, we defined good outcomes as cure or treatment completion and poor outcomes as death or treatment failure, excluding patients lost to follow-up. Patients categorized as “clinical default” (i.e., those whose treatment course deviated from standard programmatic management due to clinical decisions and whose outcomes could not be reliably classified using WHO definitions) were excluded from these analyses.

Categorization of individual-level variables

We categorized participants into three age groups (15–30, 31–45, and 46–60 years). We derived a household socioeconomic status (SES) score using principal component analysis of housing quality, water supply, and sanitation variables and categorized the resulting score into tertiles representing low, middle, and high SES³¹. Index patients were classified by duration of cough prior to diagnosis (<2 weeks, 2–4 weeks, 4–8 weeks, and >8 weeks) and by drug-resistance profile (drug-susceptible, drug-resistant, or multidrug-resistant TB).

Prevalence of TB infection status of HHCs at enrollment

We classified HHCs as TB-infected at enrollment if they reported a prior TB diagnosis, reported a positive TST within the preceding 6 months, had a TST induration >10 mm, had already been diagnosed with TB, or were referred for TB symptoms at enrollment and diagnosed with TB.

Statistical analysis

Descriptive analyses

For descriptive comparisons of baseline characteristics by incarceration history (Tables 1 and 4), continuous variables (e.g., age, body mass index) were categorized prior to analysis, and all comparisons were assessed using two-sided chi-square tests.

Index patients

We excluded index patients older than 60 years at enrollment, as none reported a history of incarceration. We evaluated the internal consistency of the TB severity score using Cronbach’s alpha, reporting the overall alpha coefficient with 95% confidence intervals and the average inter-item correlation, and assessed the contribution of individual items using item–total correlations and leave-one-item-out analyses. We evaluated the association between incarceration history and TB severity score at enrollment using linear regression with robust variance estimation, in univariable and multivariable models adjusting for age, sex, socioeconomic status, diabetes mellitus, and HIV status. Because delayed diagnosis could plausibly mediate this association, we fit an additional model further adjusting for cough duration prior to diagnosis. To facilitate scale-free comparison of severity scores between groups, we additionally summarized unadjusted differences using standardized mean differences, calculated as the difference in group means divided by the pooled standard deviation. Given the substantial contribution of BMI to the severity score, we conducted sensitivity analyses using a modified five-point score excluding BMI, and calculated standardized mean differences for both score specifications.

We used modified Poisson regression with robust variance estimation to estimate associations between incarceration history and TB treatment outcomes. We first fit univariable models, followed by multivariable models adjusting for age, sex, smoking and alcohol use, SES, prior TB, diabetes status, and HIV status. To explore potential mediation, we fit additional models further adjusting for drug-resistance profile and markers of diagnostic delay, including cough duration prior to diagnosis, sputum smear status, and cavitary disease. We conducted all analyses separately for the two TB treatment outcome definitions described above.

We examined the interval from the patient’s reported date of release from prison to the date of TB diagnosis to assess whether TB disease among individuals with a history of incarceration was plausibly linked to undiagnosed TB during incarceration or to infection acquired in prison with progression to disease after release. Because the risk of progression from infection to active disease is highest within the first 24 months after infection³², we hypothesized that, if incarceration contributed to TB risk, diagnoses would cluster shortly after release. We summarized diagnosis timing using the median time to TB diagnosis and estimated uncertainty using bias-corrected and accelerated bootstrap confidence intervals based on 2000 resamples. We further characterized the distribution of diagnosis times using the empirical cumulative distribution function (ECDF) over a 60-month period to evaluate whether diagnoses accumulated disproportionately early after release. As a complementary inferential approach, we used the Kolmogorov–Smirnov test³³ to compare the observed distribution with a uniform distribution over 0–60 months, which would be expected if incarceration had no influence on the timing of disease onset. To account for the possibility that individuals incarcerated for extended periods may have been infected early during incarceration and therefore exhibit a more constant post-release risk of progression, we conducted a sensitivity analysis restricted to individuals incarcerated for 12 months or less. As an additional plausibility check, we further evaluated diagnosis timing among individuals with extremely short incarceration durations (≤1 month), for whom meaningful prison-based exposure would be limited, to assess whether the post-release clustering of diagnoses attenuated as expected with minimal exposure.

Household contacts

We excluded HHCs younger than 15 years or older than 60 years at enrollment, as incarceration was uncommon in these age groups. We evaluated associations between (1) HHC incarceration history and (2) index patient incarceration history and the prevalence of TB infection among HHCs at enrollment using modified Poisson generalized estimating equations to account for clustering at the household level. We used a sandwich variance estimator and a working correlation matrix to obtain robust inference. We first fit univariable models, followed by multivariable models adjusting for index patient and HHC characteristics identified a priori as potential confounders of TB infection risk. These included index patient age, sex, smoking and alcohol use, sputum smear grade, duration of cough prior to diagnosis, and presence of cavitary disease, as well as HHC age, sex, Bacillus Calmette–Guérin (BCG) vaccination history, and socioeconomic status.

Given the high burden of TB exposure in prison settings, we evaluated whether even relatively short periods of incarceration were associated with increased TB infection risk among HHCs. To examine this, we conducted additional analyses in which we categorized HHC incarceration history by duration in two ways: no incarceration, ≤6 months, or >6 months; and no incarceration, ≤12 months, or >12 months.

We conducted complete-case analyses for multivariable models; fewer than 20% of observations had missing data.

We did not disaggregate all the results of this work by sex due to the very small number of female participants reporting an incarceration history.

All analyses were performed using R (version 4.5.2; R Foundation for Statistical Computing). Generalized estimating equations were fit using the geepack package; bootstrap confidence intervals were computed using the boot package. An illustrative example of the R code used for the primary analyses is provided in Supplementary Information.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.B.B., M.B.M., M.C.B., and C.-C.H. conceptualized the study and designed the methodology. J.J. was responsible for project administration. C.C.C. and R.M.Y. led the study implementation and data collection, and L.W.L. provided supervision. R.I.C. oversaw the clinical laboratory input. Z.Z. oversaw the data management and data quality assurance. C.-C.H. performed the analysis; M.B.M., C.-C.H., and A.E.M. wrote the first draft of the manuscript. C.-C.H. and Z.Z. accessed and verified all the data. All authors revised critically for interpretation and content. All authors had final responsibility for the decision to submit for publication.

Peer review

Peer review information

Nature Communications thanks Julio Croda, Helen Jack and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The de-identified, aggregate source data underlying the figures and tables in this study are provided with this paper. The individual-level participant data analyzed in this study contain sensitive personal health information, including data on tuberculosis disease, HIV status, and incarceration history, and are therefore subject to ethical and legal restrictions to protect participant privacy and confidentiality, in accordance with the conditions of the original informed consent and the approvals granted by the Institutional Review Board of Harvard Medical School and the Comité Institucional de Ética en Investigación del Instituto Nacional de Salud (CIEI-INS), Peru. The de-identified individual-level dataset is therefore not publicly available. Access to the de-identified individual-level dataset may be requested for non-commercial academic research purposes by qualified researchers affiliated with a recognized academic or research institution. Requests should be directed to the corresponding author, Megan Murray (megan_murray@hms.harvard.edu). Each request will be reviewed by the study investigators and will require (i) a brief research proposal describing the proposed use of the data, (ii) documentation of approval, or determination of exemption, from the requester’s institutional review board, and (iii) execution of a data use agreement that prohibits attempts to re-identify participants and prohibits onward sharing of the dataset. Requests will receive a response within 8 weeks of submission. Data sharing will commence upon completion of all institutional approvals and execution of the data use agreement. The dataset will remain available for 10 years following publication of this article. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74436-8.