Survival and factors associated with mortality among people with tuberculosis in Medellín, Colombia (2018–2023): a retrospective cohort study

Abstract

Objective

To determine the survival rate and prognostic factors associated with tuberculosis (TB) mortality in Medellín between 2018 and 2023.

Design

Quantitative observational analytical study of a retrospective cohort.

Setting

Based on notifications made to the Public Health Surveillance System and managed by the Secretary of Health of Medellín—Colombia between 2018 and 2023.

Participants

A total of 11 202 individuals diagnosed with TB, aged between 1 and 103 years.

Primary and secondary outcome measures

The Kaplan-Meier method was employed to determine survival and risk functions, as well as median survival. Crude HRs and adjusted HRs (aHRs) were estimated using Cox proportional hazards regression models.

Results

A median overall survival of 1410 days (3.86 years) and an adjusted mortality rate of 40 cases per 100 000 population were estimated for the study period (6 years). Factors associated with TB mortality were age (>59 years) (aHR 5.53; 95% CI 3.17 to 9.65), renal disease (aHR 2.98; 95% CI 2.27 to 3.90), HIV infection (aHR 2.82; 95% CI 1.39 to 3.32) and cancer (aHR 2.56; 95% CI 1.95 to 3.34).

Conclusions

TB survival is influenced by age and comorbidities, indicating the need for targeted strategies to protect high-risk groups. Strengthening comprehensive TB control through timely diagnosis, integrated management of chronic conditions and patient-centred care is essential to reduce preventable deaths. Furthermore, improving case notification and follow-up through integrated information platforms will contribute to more effective public health interventions.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study provides a comprehensive analysis of survival and factors associated with tuberculosis mortality in Medellín over a 6-year period, based on a large cohort of 11 202 patients reported to the public health surveillance system.

• The robust size of the cohort reinforces the accuracy and statistical power of the findings, which can be generalised to similar urban contexts in Colombia and, potentially, at the regional level in Latin America.

• Rigorous statistical methods, such as survival analysis by the Kaplan-Meier method and Cox proportional hazards models, were employed, finding risk factors associated with mortality.

• There are some inherent limitations of studies based on mass registration systems, such as the limited number of variables analysed, restricted follow-up periods and lack of integration in patient follow-up information.

Introduction

Tuberculosis (TB) is a communicable disease caused by Mycobacterium tuberculosis complex, spread when people sick with TB shed bacteria into the air, such as by coughing. It is ranked among the top 15 causes of death worldwide and, except for the COVID-19 period, TB has been the leading cause of death from a single infectious agent, ahead of HIV since 2012.¹

TB disease is often associated with poverty, vulnerability and marginalisation, leading to stigma and discrimination against those affected. The number of people acquiring the infection and developing the disease and, in turn, the number of deaths can also be reduced through multisectoral actions to address the social determinants of TB.^{1 2}

A total of 10.8 million people fell ill with TB in 2023 worldwide and 1.25 million died. In the Americas region, 309 000 TB cases were reported with a case fatality rate of 7%. In addition, an incidence rate of 28 cases per 100 000 population has been estimated.1,4

In Colombia, 14 470 TB cases were reported to the National Public Health Surveillance System (SIVIGILA) in 2021, of which 8.1% corresponded to previously treated cases and 2.4% to drug-resistant TB. In this year, the incidence was 26.53 cases per 100 000 inhabitants. Pulmonary forms accounted for 84.6% and 1737 cases of TB-HIV coinfection (12.1%) and 1347 TB-COVID-19 (9.4%) were reported. In terms of mortality, a rate of 2.14 cases per 100 000 population and a case fatality rate of 7.8% were reported, which corresponds to about 1131 deaths.^{2 3}

In Medellín, the incidence of TB between 2014 and 2018 remained between 50 and 60 cases per 100 000 inhabitants, double the national average. Territorial analysis shows the highest concentration of cases in areas of low socioeconomic strata and some marginalised areas.^{5 6}

Despite global progress in TB prevention and treatment, mortality remains high, particularly in low-income and middle-income countries where health system challenges and social inequities persist, making it crucial to understand the patterns and factors that influence survival. In Latin America, case fatality rates exceed global targets, and TB continues to disproportionately affect vulnerable populations. However, evidence on survival patterns and determinants of mortality at the local level is limited, particularly in urban settings.⁷ Generating this type of evidence is essential to inform global TB control strategies, reduce preventable deaths and guide tailored interventions in high-burden contexts.

The study was conducted in Medellín, as it is one of the few cities in Colombia with a long-standing, municipality-led TB control programme that has ensured one of the highest levels of data quality in the country. This context makes the city a suitable setting for analysing TB mortality with reliability. Nevertheless, despite the programme’s long trajectory, there is a lack of evidence on survival analysis and the factors influencing mortality in this population. The present study seeks to address this gap by determining survival and associated factors of TB mortality in Medellín between 2018 and 2023, positioning the city as a representative setting.

Methods

A quantitative observational analytical study of a retrospective cohort of 11 202 individuals was conducted, based on notifications made to SIVIGILA (National Surveillance System)⁸ and managed by the Medellín TB control programme between 2018 and 2023. This study relied on secondary data systematically collected and standardised by the national and local health authorities through the surveillance system. Data extraction was performed using standardised templates provided by the local health authority. Records were reviewed for completeness and consistency.

The study population included all individuals with a TB diagnosis residing in Medellín during the specified period. The unit of observation was the totality of the records corresponding to this period and the unit of analysis was the person registered.

The primary outcome was TB-specific mortality, defined as death attributable to TB. Time-to-event (in days) was calculated from the date of symptom onset to the date of death or, for censored cases, to the date of last follow-up. The cause of death was certified by physicians, coded according to the International Classification of Diseases-10, and officially registered in the SIVIGILA, ensuring accurate attribution to TB rather than secondary causes.

Sample size and power considerations

This study included the total number of TB cases reported in Medellín between 2018 and 2023 (n=11 202). As TB mortality is a mandatory notification event in Colombia, under-reporting is expected to be minimal. Nevertheless, even under a hypothetical scenario of 10% underreporting (n ≈ 12 447), power calculations indicated a statistical power greater than 99%.

Study variables

All variables were defined according to the criteria established by the Colombian Ministry of Health and SIVIGILA protocols, which provide standardised operational definitions for surveillance purposes. A total of 26 variables were analysed, covering sociodemographic, clinical, contextual and treatment-related characteristics, as well as the outcome of interest.

The variable ‘failure’ was defined as the occurrence of death (yes/no) during the follow-up period and was used in the survival analysis as the event of interest. This variable was accompanied by the time to event, measured in days from the date of TB diagnosis to the date of death or censoring (end of follow-up).

Sociodemographic variables include sex, age, socioeconomic stratum and social security affiliation; clinical variables included comorbidities such as HIV infection, diabetes, renal disease and cancer; and contextual factors included migrant status, displacement, homelessness, incarceration and healthcare worker status. Nutritional status was classified into underweight (<18.5 kg/m²), normal (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²) and obesity (≥30.0 kg/m²).⁹ For analytical purposes, an additional binary variable was created for underweight, where ‘yes’ corresponds to body mass index (BMI)<18.5 kg/m² and ‘no’ corresponds to BMI ≥18.5 kg/m². Treatment-related variables included TB type, hospitalisation and treatment history.

For the definition of the COVID-19 pandemic periods in Colombia, the prepandemic period was considered to be cases reported during 2018 and 2019, and the pandemic period was defined as the time between 1 January 2020 and 31 December 2022.¹⁰ The postpandemic was defined for the year 2023, according to the guidelines of the Colombian presidency.¹¹ However, for the calculation of HRs the variable was subsequently dichotomised as pandemic (yes), defined as 1 January 2020–31 December 2022, and non-pandemic (no), defined as the periods before 1 January 2020 and after 31 December 2022.

Treatment history was classified into two categories. A ‘new case’ referred to an individual who was diagnosed with TB for the first time during the study period and had no previous history of TB treatment. A ‘previously treated case’ referred to an individual with a documented history of TB and treatment in earlier years, who re-entered the care pathway after being newly diagnosed during the study period.

The variable ‘persons with disability’ was defined according to the Colombian national guideline as persons with medium-term or long-term physical, mental, intellectual or sensory impairments which, in interaction with various barriers, including attitudinal barriers, may hinder their full and effective participation in society on an equal basis with others.¹² The variable ‘displacement’ was defined as any individual who has been forced to migrate within the national territory, abandoning their place of residence or habitual economic activities, because their life, physical integrity, personal safety or freedom has been violated or is under direct threat.¹³

In the Colombian health system, the contributory scheme refers to formal workers and their families who finance their affiliation through payroll-based contributions, while the subsidised scheme is financed by the State and covers individuals in conditions of poverty or vulnerability. The uninsured population corresponds to those without effective affiliation. For analytical purposes, the uninsured and the subsidised groups were combined, as both reflect disadvantaged socioeconomic conditions and limited access to health services.¹⁴

In Colombia, socioeconomic status is officially classified into six strata based on the physical characteristics of the dwelling and its surroundings rather than directly on household income. Stratum 1 corresponds to the lowest living conditions and stratum 6 to the highest, with strata 2–3 considered low to lower-middle, 4–5 middle to upper-middle and 6 high socioeconomic status. This categorisation is widely applied for the allocation of public service subsidies and is commonly used as a proxy for household living conditions.¹⁵

Data analysis

To describe the characteristics of the cohort, a descriptive univariate and bivariate statistical analysis was initially performed. Subsequently, univariate analysis of the information was performed using the survival function with the Kaplan-Meier method.

In the bivariate analysis of the information, we cross-referenced the variable failure with sociodemographic and clinical variables. The Wald test was performed on these crosses, and the HR was calculated, using proportional hazards regression models or Cox regression, which considers, in addition to the occurrence of the event, the time it takes to occur. The goodness of fit of these simple models was assessed by the Akaike information criterion (AIC).¹⁶

Data are presented using Kaplan-Meier estimator curves. The test of equality of log-rank survival distributions was used for the different levels of categorical variables. Point estimates and 95% confidence intervals were obtained for the mean survival time of all individuals and the HR.

Crude mortality rates were calculated by dividing the number of TB-related deaths by the total study population and expressed per 100 000 inhabitants. Age-specific and sex-specific mortality rates were estimated for different subgroups of the population.

Expected deaths were obtained by applying the age-specific and sex-specific mortality rates from the WHO standard population¹⁷ to our study cohort. Specific and adjusted mortality rates were calculated taking into account both the WHO standard population and the population of Medellín at the mid-term of the cohort (2020), reported by the National Administrative Department of Statistics (DANE) according to the 2018 National Population and Housing Census projections.¹⁸ For the calculation of mortality rates, the population of Medellín in 2020 was used as the reference, estimated at approximately 2.5 million inhabitants according to DANE. The city’s population has followed a stable growth trend in recent years, with no abrupt variations.

Although the COVID-19 pandemic in 2020 led to an increase in deaths, this effect was proportionally small relative to the total population and therefore did not substantially modify the denominator used for the estimation of mortality rates. Adjusted mortality rates were then calculated using the direct standardisation method.

Finally, multivariate Cox models were developed considering the variables that presented a statistically significant association (p<0.05) in the simple models and in order of AIC (lowest to highest AIC), and that presented biological and clinical plausibility, thus constructing a multivariate Cox proportional hazards model. The proportional hazards assumption was verified by analysing the Schoenfeld residuals.

SPSS statistical software V.28 (licensed by the University of Antioquia) and R software were used.

Patient and public involvement

Patients were not directly involved in this research. However, the public was represented through the participation of the municipal TB control programme and the health professionals from the Medellín Secretariat of Health, who contributed to the design of the study, the provision of information and the interpretation of the results. In addition, a meeting was held with the Secretariat of Health, through the TB programme, to present and discuss the findings. The results provide evidence on survival and mortality determinants that contribute to strengthening local TB prevention and care strategies, supporting decision-making for improving surveillance, resource allocation and clinical management within the programme.

Results

General characteristics of the cohort

A total of 11 202 SIVIGILA records were analysed. The cohort corresponds to people diagnosed with pulmonary and/or extrapulmonary TB by one or more methods such as smear microscopy (82.8%), microbiological culture (57.8%) and molecular testing (46.8%), as well as diagnosis confirmed by clinical condition (97.5%), epidemiological criterion (26.7%), radiological test (70%) and adenosine deaminase test (7.7%). Within this time frame, 129 individuals were identified who had a previous history of TB and treatment in earlier years.

The follow-up time of individuals with TB had a mean of 301.6 days and a median of 278 days, with a maximum of 1967 days (5.3 years). Among those who died, the median follow-up was 62 days (IQR: 29–132; range: 1–1705 days) from symptom onset to death. This variable was not normally distributed (Shapiro-Wilk test p<0.01).

The cohort population is predominantly male (62.5%). There is a predominance of men between the ages of 20 and 40. The age group that includes the largest population is between 21 and 30 years old. Age has a median of 39 years, a range of 102 years (between 1 and 103 years) and an IQR of 30 years and does not have a normal distribution (Shapiro-Wilk test p<0.01).

Regarding the healthcare system and social conditions, 38.9% of the people belong to the subsidised healthcare regime, which covers low-income and vulnerable populations, and 4.7% are not affiliated to the system. Meanwhile, 55.3% are in the contributory, special or exceptional regimes. No insurance information was obtained for 122 individuals (1.1%). In terms of socioeconomic classification, 68.6% belong to strata 1 and 2 (the lowest levels) with 37.9% of the population in stratum 2.

During the study period, 1179 deaths were reported, with a case fatality rate of 10.5%. In the prepandemic period for COVID-19, 3363 cases were reported, 9.2% of whom died, then 5337 cases were reported during the pandemic, 12.7% of whom died (n=680), and 2502 cases were reported in the postpandemic period, with a case fatality rate of 7.6% (n=191).

The year 2020 had the highest case fatality rate with 15.7% (95% CI 13.9% to 17.6%) (n=234). Differences between years were significant (X², p<0.001). Table 1 shows the behaviour of the case fatality rate for each year. The case fatality for 2020 is significantly higher.

Table 1. Tuberculosis case fatality by year and pandemic period in Medellín, 2018–2023.

Variable	Categories	Cases TB	Deceased (n)	Fatality (%)	95% CI fatality
					Lower limit (%)	Upper limit (%)
Year of notification	2018	1531	144	9.4	8.0	10.9
	2019	1832	164	9.0	7.7	10.3
	2020	1488	234	15.7	13.9	17.6
	2021	1762	224	12.7	11.2	14.3
	2022	2087	222	10.6	9.4	12.0
	2023	2502	191	7.6	6.6	8.7
	Total	11 202	1179	10.5	10.0	11.1
Pandemic by year of notification	Prepandemic	3363	308	9.2	8.2	10.2
	Pandemic (2020–2022)	5337	680	12.7	11.9	13.7
	Postpandemic	2502	191	7.6	6.6	8.7
	Total	11 202	1179	10.5	10.0	11.1

TB, tuberculosis.

Approximately half of the cases (48.1%) are concentrated in five districts: Manrique (1,316), Santa Cruz (718), Robledo (701), Buenos Aires (683) and Aranjuez (661). The case fatality rate in these places where the highest number of cases occurred was between 9.4% and 11.3%. Among the population that died, the highest proportion was observed among street dwellers, with 11.8%, followed by residents of the Manrique district, at 11.3%.

The cohort included comorbidities such as HIV infection (16.0%), diabetes (8.3%), chronic obstructive pulmonary disease (5.8%), cancer (3.5%), renal disease (3.2%), rheumatoid arthritis (1.6%), liver disease (0.6%) and silicosis (0.2%). Individuals with liver disease had the highest proportion of deaths; in this case, of the 66 people with this condition, 19 died (28.8%). This was followed by those with kidney disease (356 people) with 26.1% of deaths.

Survival analysis

The mean overall survival for the cohort was 3.86 years (1410 days, 95% CI 1324 to 1497, SD 44 days). Half of the population had a survival of 4.34 years or less (1583 days—median of survival).

The mean survival was estimated with the log-rank test that compares probability distributions of survival between variable categories.

Table 2 shows the mean survival time of people with TB according to different characteristics. The most pronounced gaps were observed in hospitalisation status, HIV infection and renal disease. Patients hospitalised during treatment had a mean survival of 1140 days, compared with 1655 days among those not hospitalised, which represents a reduction of more than 500 days of life expectancy after diagnosis. HIV coinfection was also linked to a markedly shorter survival, with individuals surviving on average 1040 days compared with 1534 days in those without HIV, a difference of nearly 1.5 years. Similarly, individuals with renal disease had a mean survival of 1177 days compared with 1417 days among those without this comorbidity, representing a reduction of approximately 240 days (8 months). All comparisons were statistically significant (log Rank test, p<0.01).

Table 2. Mean survival of people with tuberculosis in Medellín, 2018–2023.

Variable	Categories	Average survival rate (in days)				Log rank
		Estimate	SD	95% CI
				Lower limit	Upper limit	X2	P value
Hospitalised patient	Yes	1140	42	1058	1223	584.72	<0.01
	No	1655	60	1538	1773
	Overall	1410	44	1324	1497	---
Age	<19	1352	173	1012	1691	305.72	<0.01
	Youth (19–26)	1431	63	1308	1555
	Adulthood (27–59)	1385	47	1293	1476
	Older person (>59)	1494	35	1425	1564
	Overall	1410	44	1324	1497	---
HIV infection	Yes	1040	49	943	1137	211.74	<0.01
	No	1534	53	1430	1639
	Overall	1410	44	1324	1497	---
Social security	Uninsured or subsidised scheme	1219	35	1151	1287	119.96	<0.01
	Contributory	1667	55	1560	1775
	Overall	1397	45	1309	1485	---
Renal disease	Yes	933	35	864	1003	107.62	<0.01
	No	1417	44	1330	1505
	Overall	1410	44	1324	1497	---
Cancer	Yes	1281	61	1161	1401	95.25	<0.01
	No	1398	46	1308	1488
	Overall	1401	45	1313	1488	---
Nutritional status	Underweight	1185	40	1106	1263	87.28	<0.01
	Normality	1491	59	1377	1606
	Overweight	1229	11	1208	1251
	Obesity	1498	107	1287	1708
	Overall	1412	46	1321	1502	---
Underweight*	Yes	1185	40	1106	1263	81.61	<0.01
	No	1511	53	1407	1614
	Overall	1412	46	1321	1502	---
COPD	Yes	990	24	942	1037	74.81	<0.01
	No	1408	45	1319	1497
	Overall	1401	45	1313	1488	---
Sex	Man	1294	36	1223	1364	59.58	<0.01
	Woman	1476	69	1340	1612
	Overall	1410	44	1324	1497	---
Homeless person	Yes	1231	61	1113	1350	47.64	<0.01
	No	1408	52	1306	1510
	Overall	1411	44	1324	1497	---
Diabetes	Yes	1081	38	1006	1156	15.09	<0.01
	No	1421	45	1333	1510
	Overall	1410	44	1324	1497	---
Person with disability	Yes	607	34	541	673	5.01	0.03
	No	1412	44	1325	1498
	Overall	1411	44	1324	1497	---
Silicosis	Yes	724	69	589	858	2.94	0.09
	No	1411	44	1324	1497
	Overall	1410	44	1324	1497	---
Treatment history	Previously treated case	1389	61	1270	1509	0.06	0.81
	New case	1398	59	1283	1513
	Overall	1410	44	1324	1497	---

^*Underweight: body mass index <18.5 kg/m2.

COPD, chronic obstructive pulmonary disease.

Mortality rates

The trend in mortality rates for the entire study period is different according to age and sex, with the specific mortality rate increasing with age. In the 25–29 and 40–49 age groups, rates between 45.1 and 54.9 cases per 100 000 inhabitants were found, while from the age of 65 onwards, rates between 99.6 and 180.1 cases per 100 000 inhabitants were found. An overall adjusted mortality rate of 40 cases per 100 000 population was found (see table 3). The adjusted mortality rate for men was 65.5 cases per 100 000 population, while the adjusted mortality rate for women was 19.6 per 100 000 population (rate ratio=3.3) (see table 4).

Table 3. Mortality rate by age group of people with tuberculosis in Medellín, 2018–2023.

Age (years)	Population Medellín (2020)	TB deaths	Rate specific	Deaths expected	Standard population (WHO)
0–4	144 919	4	2.8	0.2	8860
5–9	150 928	0	0.0	0.0	8690
10–14	160 587	1	0.6	0.1	8600
15–19	186 603	16	8.6	0.7	8470
20–24	228 270	54	23.7	1.9	8220
25–29	237 028	107	45.1	3.6	7930
30–34	211 886	115	54.3	4.1	7610
35–39	190 665	69	36.2	2.6	7150
40–44	162 185	89	54.9	3.6	6590
45–49	146 753	68	46.3	2.8	6040
50–54	157 154	90	57.3	3.1	5370
55–59	151 248	103	68.1	3.1	4550
60–64	125 473	115	91.7	3.4	3720
65–69	96 892	114	117.7	3.5	2960
70–74	70 312	70	99.6	2.2	2210
75–79	45 939	69	150.2	2.3	1520
>79	52 750	95	180.1	2.7	1510
Total	2 519 592	1179		40.0	100 000
Crude rate (×100 000 inhabitants)		46.8
Adjusted rate (×100 000 inhabitants)		40.0

TB, tuberculosis.

Table 4. Mortality rates by sex and age groups of people with tuberculosis in Medellín, 2018–2023.

Age (Years)	Sex								Standard population (WHO)
	Men				Women
	Population Medellín (2020)	TB deaths	Specific rate	Expected deaths	Population Medellín (2020)	TB deaths	Specific rate	Expected deaths
0–4	73 936	2	2.7	0.2	70 983	2	2.8	0.2	8860
5–9	76 783	–	–	–	74 145	–	–	–	8690
10–14	81 321	1	1.2	0.1	79 266	–	–	–	8600
15–19	94 213	7	7.4	0.6	92 390	9	9.7	0.8	8470
20–24	114 407	40	35.0	2.9	113 863	14	12.3	1.0	8220
25–29	118 712	86	72.4	5.7	118 316	21	17.7	1.4	7930
30–34	105 139	89	84.6	6.4	106 747	26	24.4	1.9	7610
35–39	91 764	59	64.3	4.6	98 901	10	10.1	0.7	7150
40–44	74 642	70	93.8	6.2	87 543	19	21.7	1.4	6590
45–49	64 909	46	70.9	4.3	81 844	22	26.9	1.6	6040
50–54	68 178	68	99.7	5.4	88 976	22	24.7	1.3	5370
55–59	64 542	75	116.2	5.3	86 706	28	32.3	1.5	4550
60–64	52 399	83	158.4	5.9	73 074	32	43.8	1.6	3720
65–69	39 678	74	186.5	5.5	57 214	40	69.9	2.1	2960
70–74	28 199	49	173.8	3.8	42 113	21	49.9	1.1	2210
75–79	17 911	45	251.2	3.8	28 028	24	85.6	1.3	1520
>79	19 463	61	313.4	4.7	33 287	34	102.1	1.5	1510
Total	1 186 196	855		65.5	1 333 396	324		19.6	100 000
Crude rate (x100 000 inhabitants)			72.1			24.3
Adjusted rate (x100 000 inhabitants)			65.5			19.6
Adjusted rate ratio				3.3

TB, tuberculosis.

Prognostic factors associated with mortality

In the bivariate analysis (table 5), individuals with TB aged over 69 years showed the strongest association with mortality (HR 8.71; 95% CI 5.29 to 14.33), being hospitalised (HR 6.19; 95% CI 5.23 to 7.32) and having renal disease (HR 2.91; 95% CI 2.36 to 3.60). Multivariate Cox models were developed with those variables that were significant and with biological and clinical plausibility to adjust the HR (see table 6).

Table 5. Factors associated with survival of people with tuberculosis in Medellín, 2018–2023.

Variable	Categories	Final condition				Wald P value	HR	95% CI of HR
		Dead		Alive				Lower limit	Upper limit
		n	%	n	%
Sex	Man	855	72.5	6142	61.3	<0.01	1.65	1.45	1.87
	Woman	324	27.5	3881	38.7			Ref.
Social security	Uninsured or subsidised scheme	691	59.7	4197	42.3	<0.01	1.91	1.70	2.15
	Contributory	466	40.3	5726	57.7			Ref.
Pandemic	Yes	680	57.7	4657	46.5	<0.01	1.50	1.34	1.69
	No	499	42.3	5366	53.5			Ref.
Stratum	1	206	24.8	1510	19.2	0.01	1.72	1.11	2.67
	2	388	46.7	3853	49.0		1.32	0.86	2.03
	3	215	25.9	2204	28.0		1.35	0.87	2.09
	4, 5, 6	22	2.6	292	3.7			Ref.
Person with disability	Yes	14	1.2	66	0.7	0.03	1.81	1.07	3.07
	No	1151	98.8	9857	99.3			Ref.
Displacement	Yes	2	0.2	40	0.4	0.22	0.42	0.11	1.70
	No	1163	99.8	9885	99.6			Ref.
Migrant	Yes	39	3.3	297	3.0	0.12	1.29	0.93	1.77
	No	1126	96.7	9634	97.0			Ref.
Individuals deprived of liberty	Yes	12	1.0	130	1.3	0.73	0.90	0.51	1.60
	No	1153	99.0	9798	98.7			Ref.
Homeless person	Yes	139	11.9	608	6.1	<0.01	1.85	1.55	2.22
	No	1026	88.1	9321	93.9			Ref.
Hospitalised patient	Yes	1022	86.7	5148	51.4	<0.01	6.19	5.23	7.32
	No	157	13.3	4875	48.6			Ref.
Type of tuberculosis	Pulmonary	959	81.3	8122	81.0	0.74	1.03	0.89	1.19
	Extrapulmonary	220	18.7	1901	19.0			Ref.
Treatment history	Previously treated case	129	10.9	859	8.6	0.81	0.98	0.81	1.18
	New case	1050	89.1	9164	91.4			Ref.
Healthcare worker	Yes	2	0.2	166	1.7	<0.01	0.10	0.03	0.41
	No	1177	99.8	9857	98.3			Ref.
Age	<19	16	1.4	669	6.7	<0.01		Ref.
	Youth (19–26)	94	8.0	1754	17.5		2.16	1.27	3.67
	Adulthood (27–59)	606	51.4	5612	56.0		4.15	2.52	6.81
	Older person (>59)	463	39.3	1988	19.8		8.71	5.29	14.33
Nutritional status*	Normality	583	50.1	5355	57.2	<0.01		Ref.
	Underweight	431	37.1	2275	24.3		1.64	1.45	1.86
	Overweight	111	9.5	1373	14.7		0.77	0.62	0.94
	Obesity	38	3.3	351	3.8		0.95	0.68	1.32
Underweight	Yes	431	37.1	2275	24.3	<0.01	1.72	1.53	1.94
	No	732	62.9	7079	75.7			Ref.
HIV infection	Yes	363	30.8	1424	14.2	<0.01	2.43	2.15	2.75
	No	816	69.2	8599	85.8			Ref.
COPD	Yes	129	11.2	507	5.2	<0.01	2.20	1.83	2.64
	No	1024	88.8	9222	94.8			Ref.
Silicosis	Yes	6	0.5	22	0.2	0.09	1.99	0.89	4.44
	No	1173	99.5	10 001	99.8			Ref.
Diabetes	Yes	132	11.2	798	8.0	<0.01	1.43	1.19	1.71
	No	1047	88.8	9225	92.0			Ref.
Renal disease	Yes	93	7.9	263	2.6	<0.01	2.91	2.36	3.60
	No	1086	92.1	9760	97.4			Ref.
Hepatic disease	Yes	19	1.6	47	0.5	<0.01	2.83	1.80	4.45
	No	1134	98.4	9682	99.5			Ref.
Cancer	Yes	95	8.2	288	3.0	<0.01	2.73	2.21	3.37
	No	1058	91.8	9441	97.0			Ref.
Rheumatoid arthritis	Yes	27	2.3	144	1.5	0.03	1.53	1.05	2.24
	No	1126	97.7	9585	98.5			Ref.

^*Normality: 18.5–24.9 kg/m2. Underweight: <18.5 kg/m2. Overweight: 25.0–29.9 kg/m2. Obesity: ≥30.0 kg/m2.

COPD, chronic obstructive pulmonary disease.

Table 6. Multivariate Cox regression model of factors associated with survival of people with tuberculosis in Medellín, 2018–2023.

Feature	B	Wald	P value	HR	95% CI of HR
					Lower limit	Upper limit
Youth (19–26)	0.25	0.68	0.41	1.29	0.71	2.35
Adulthood (27–59)	0.75	6.92	0.01	2.11	1.21	3.69
Older person (>59)	1.71	36.29	<0.01	5.53	3.17	9.65
Renal disease	1.09	62.61	<0.01	2.98	2.27	3.90
HIV infection	1.04	154.40	<0.01	2.82	2.39	3.32
Cancer	0.94	46.78	<0.01	2.56	1.95	3.34
Person with disability	0.69	4.72	0.03	2.00	1.07	3.74
Social security (Uninsured or subsidised scheme)	0.42	28.60	<0.01	1.52	1.30	1.76
Male sex	0.32	15.35	<0.01	1.38	1.17	1.62
Status as a homeless person	0.33	8.02	<0.01	1.39	1.11	1.74
Underweight (BMI <18.5 kg/m)2	0.45	35.92	<0.01	1.57	1.35	1.82

BMI, body mass index; Β, beta coefficient (β).

It was found that individuals aged 27–59 years had more than twice the hazard of death compared with those under 19 years (adjusted HR, aHR 2.11; 95% CI 1.21 to 3.69). In the group aged over 59 years, the effect was even stronger, with more than a fivefold increased hazard of death (aHR 5.53; 95% CI 3.17 to 9.65).

The second determinant of survival was renal disease, with individuals affected having nearly a threefold higher hazard of death compared with those without the condition (aHR 2.98; 95% CI 2.27 to 3.90). The third factor was HIV infection, which conferred approximately a threefold higher hazard of death compared with individuals without HIV infection (aHR 2.82; 95% CI 1.39 to 3.32).

The last factors that entered the Cox regression model were having cancer (aHR 2.56; 95% CI 1.95 to 3.34), having a disability (aHR 2.00; 95% CI 1.07 to 3.74), affiliation to the subsidised scheme or no affiliation to the health system (aHR 1.52; 95% CI 1.30 to 1.76), being homeless (aHR 1.39; 95% CI 1.11 to 1.74) and having a BMI of less than 18.5 kg/m² (aHR 1.57; 95% CI 1.35 to 1.82). Men had a 38% higher hazard of death compared with women (aHR 1.38; 95% CI 1.17 to 1.62). The proportion of deaths in women was 7.7% (324/4205) while in men it was 12.2% (855/6997), (58.4% higher).

Figure 1A represents the survival function from symptom onset without accounting for covariates, while figure 1B shows the survival function adjusted for covariates using the Cox regression model. The median survival (the time at which 50% of individuals have experienced the event) was 1583 days for the unadjusted curve and 1625 days for the adjusted curve. A progressive decline in survival is observed over time, with a steeper decrease within the first 125 days and a more gradual slope after approximately 250 days. The slightly longer median survival in the adjusted curve indicates that accounting for covariates modestly shifts survival estimates, reflecting the influence of prognostic factors. A sharp drop in survival is evident around 1500 days in both curves.

Figure 1. Crude and adjusted survival function on the mean of Cox regression covariates of people with tuberculosis in Medellín, 2018–2023. (A) Crude survival function, (B) adjusted survival function.

Online supplemental figure 2 illustrates the survival functions for each of the variables included in the Cox regression model, which provide greater predictive power for the survival of individuals with TB, stratified by various sociodemographic and clinical characteristics. Survival outcomes vary significantly across groups, with older individuals (>59 years), those with renal disease, HIV, cancer, disabilities or underweight status (BMI <18.5 kg/m²) exhibiting notably poorer survival. Similarly, individuals with subsidised or no health insurance and those living on the streets experience reduced survival rates.

Discussion

This study analysed the survival of a cohort of individuals with TB in Medellín. The median survival observed (3.86 years) was comparable to that reported in a Danish cohort of TB individuals aged 18 years or older notified between 1990 and 2018.¹⁹ Although these populations differ in terms of socioeconomic context and health system characteristics, the similarity in survival suggests that TB continues to impose a substantial long-term burden even in countries with high-income health systems. This comparative finding reinforces the notion that survival after TB diagnosis remains limited across diverse settings, and that comorbidities and social determinants are critical drivers of outcomes.

Higher case fatality was observed during the COVID-19 pandemic. In 2020 it increased to 15.7%, even though the 95% CI of that estimate indicates that the case fatality may have been as high as 17.6%, possibly due to the disruption of health services, the prioritised care of the pandemic and the increased vulnerability of people with TB to respiratory infections. COVID-19 was not recorded in the data source used, but other studies reported case fatality rates of 23.5% in TB-COVID-19 coinfection, suggesting that these individuals are more likely to suffer severe disease and die.²⁰

During the first year of the COVID-19 pandemic, many healthcare systems worldwide became saturated, and services for other diseases were deprioritised. This situation led to an increase in morbidity and mortality indirectly related to COVID-19. In several countries, including Colombia, access to TB diagnostic and treatment services was reduced, contributing to higher case fatality rates, increased transmission and decreased reporting or potential under-reporting of TB cases.21,23

Age emerges as the most influential factor in the survival of TB individuals, with a significantly higher risk observed in older age groups. Individuals over 59 years had a 5.53-fold higher hazard of death compared with those under 19 years, highlighting the need for specific approaches to TB management in older adults. As emphasised by Vishnu Sharma et al,²⁴ older people often do not present with the classic symptoms, clinical and radiological signs of TB, which may result in misdiagnosis or delayed diagnosis. In addition, the presence of multiple diseases makes treatment difficult.

The presence of comorbidities such as renal disease is associated with a higher risk of earlier mortality (aHR 2.98). These individuals are a high-risk group and their treatment requires adjustment due to renal elimination of some medications and complications arising from their treatment.²⁵

HIV infection (aHR: 2.82) and cancer (aHR: 2.56) are associated with decreased survival time in TB individuals. To provide context, similar studies have reported comparable associations: the Danish cohort¹⁹ (1990–2018) found an aHR of 2.41 for HIV infection, while a Chinese cohort²⁶ reported an aHR of 3.51. Comparing these results across different populations and healthcare settings helps to assess the consistency and generalisability of our findings, having HIV as a strong predictor of decreased survival in diverse epidemiological contexts. These comparisons also underscore the importance of integrated care strategies for TB individuals with comorbidities, as early identification and management of conditions such as HIV and cancer can influence survival outcomes.

Individuals belonging to the subsidised regime or without affiliation to the healthcare system presented an increased risk of mortality, reflecting the barriers in access to quality health services. In addition, street dwellers and those with underweight also show increased vulnerability, similarly reported by Lumu et al²⁷ who found an aHR of 3.79 in individuals presenting with HIV-TB and underweight.

There is a majority participation of people belonging to low socioeconomic strata (68.6% in strata 1 and 2), highlighting this vulnerability as one of the social determinants of health.²⁸ These demographic characteristics underline the importance of addressing social inequalities in health as an integral part of TB control and treatment strategies.²⁹

The survival gaps identified in this study, particularly those related to age, hospitalisation and comorbidities such as HIV and renal disease, reflect not only the biological severity of TB but also structural barriers that influence patient outcomes. Limited access to timely diagnosis, fragmented management of comorbidities and inequities in continuity of care may contribute to premature mortality. Addressing these gaps requires the implementation of mechanisms that ensure the provision of healthcare services in a comprehensive, accessible and equitable manner, overcoming economic, geographic and social barriers that hinder access to effective TB care.³⁰

The study was conducted in Medellín, one of the few cities in Colombia with a long-standing municipality-led TB control programme that guarantees high-quality data. The findings of this study are particularly relevant for public health in urban contexts such as Medellín and may be generalisable to other settings with comparable characteristics (as reported in previous studies^{19 26 27 31}). Medellín shares epidemiological and social conditions with other large cities where TB transmission is strongly influenced by socioeconomic inequalities and healthcare system limitations, which reinforces the external validity of our results.

The observed case fatality and the significant association of comorbidities underscore the need to strengthen healthcare systems and TB control programmes to improve continuity and quality of care, as well as timely diagnosis.

Hepatotoxicity induced by anti-TB drugs, such as isoniazid, rifampicin and pyrazinamide, is a challenge in TB management. People with liver failure face an elevated risk of severe complications during treatment, which directly impacts their prognosis.³² This population requires continuous follow-up, with adjustments to treatment regimens and strict monitoring of liver function, as adverse effects can accelerate progression to organ failure, aggravating the medical condition, where it is essential to articulate multidisciplinary strategies from hepatology and pharmacovigilance to mitigate these risks and improve patient survival.³³

Strengths and limitations of the study

The strengths of the study include the large sample size of 11 202 individuals, the use of a well-established public health surveillance database, and the application of rigorous statistical methods such as Kaplan-Meier survival analysis and Cox proportional hazards models to identify prognostic factors.

The study contributes to the body of knowledge by providing detailed estimates of TB survival and identifying key risk factors in a large urban cohort in Medellín. The findings can inform clinical practice by guiding healthcare professionals to prioritise monitoring and management of individuals with high-risk characteristics, such as older age, renal disease, HIV infection and cancer. At the policy level, the results support the development of targeted interventions, resource allocation and the strengthening of health information systems to improve TB outcomes.

In the present study, the statistical power provided by the number of people in the cohort and the precision of the estimates made stands out. However, there are some limitations inherent to the nature of studies based on massive registration systems, such as the small number of variables studied, limited follow-up times and disarticulation of patient follow-up information.^{19 26 27 31}

The Colombian SIVIGILA system is outdated and disjointed in relation to patient follow-up and the recording of complementary information (filled out by different health institutions and managed by the Ministry of Health) such as susceptibility tests, adverse effects, comorbidities, drug supply and permanence in the programme. This information is consolidated independently and outside the national system, which makes it impossible to have all the updated and standardised information for each person on the same digital platform.

Real-world data is fundamental to improving the understanding and management of diseases such as TB, as it can capture information on the effectiveness of treatment in routine care settings where patient conditions and characteristics are more heterogeneous.^{34 35} However,making full use of these data requires an efficient epidemiological surveillance system that is up to date and centralised. In Colombia, the Public Health Surveillance System (SIVIGILA) still has limitations in the integration of key information, such as comorbidity records and treatment adherence, which hinders comprehensive follow-up of TB patients. Optimising these systems, as recommended by the WHO, is essential for a faster and more effective response to public health needs. This would not only improve individual case management, but also facilitate continuous, evidence-based evaluation for decision-making.³⁶

The use of the WHO standard population, also adopted by the Colombian Ministry of Health, allowed for the standardisation of TB mortality rates in Medellín and ensured comparability with both national and international reports. This methodology, based on population structure rather than individual follow-up, does not incorporate time-to-event data, but it is widely accepted in public health surveillance to describe and compare mortality patterns. Our complementary use of survival analysis (Kaplan-Meier and Cox models) addressed the time-at-risk dimension, providing a comprehensive assessment of TB mortality.

Another limitation of this study is the potential misclassification bias related to the definition of the pandemic period. Although national guidelines defined 2020–2022 as the pandemic years, the first confirmed COVID-19 case in Colombia occurred in March 2020. Therefore, the first months of 2020 may represent a ‘grey period’ in which some cases and deaths could have been misclassified. However, given the population size and the overall distribution of cases, the impact on the results is likely minimal.

Conclusions

This study estimated a median survival of 1410 days (approximately 3.86 years) among individuals diagnosed with TB in Medellín between 2018 and 2023, with an adjusted mortality rate of 40 cases per 100 000 population over the study period. The analysis identified key factors associated with increased TB mortality, including age over 59 years, renal disease, HIV infection and cancer. These findings provide evidence on the magnitude of TB mortality in the city and the subgroups most at risk.

The results indicate the need for targeted interventions that address the identified prognostic factors and support timely diagnosis and treatment. Integrating strategies that consider comorbidities and vulnerable age groups can contribute to improving survival among TB individuals. Additionally, these findings highlight the relevance of strengthening health information systems to monitor outcomes and guide public health actions.

TB mortality is largely preventable with timely and appropriate interventions. Modifiable factors such as access to health services, early detection, patient follow-up and support programmes, drug supply facilities, continuity of treatment, health education and effective management of comorbidities are crucial in reducing TB mortality.

Data availability statement

Data are available on reasonable request.