Incidence of tuberculous infection in an African city with endemic tuberculosis: A population-based study

Abstract

Background:

Current metrics for tuberculosis transmission include tuberculosis notifications, disease mortality, and prevalence surveys. These metrics are helpful to national tuberculosis programs to assess the burden of disease, but they do not directly measure incident infection in the community.

Methods:

To estimate incidence of M. tuberculosis infection in Kampala, Uganda, we performed a prospective cohort study between 2014 and 2017 which enrolled of 1275 adult residents without signs of tuberculous infection (tuberculin skin test (TST) < 5 mm and no signs of tuberculosis disease) and followed them for conversion of TST at one year.

Results:

During follow-up, 194 participants converted the TST and 158 converted by one year. The incidence density of TST conversion was 13.2 conversions per 100 person-year (95% CI: 11.6, 15.1), which corresponds to an annual cumulative incidence of tuberculous infection of 12.4% (95% CI: 10.7%, 14.3%). Cumulative incidence was greater among older participants and among men. Among participants who reported prior exposure to tuberculosis cases, the cumulative risk was highest among those reporting exposure during follow-up.

Conclusions:

The high annual incidence of infection suggests that residents of Kampala have adequate contact for infection with undetected, infectious cases of tuberculosis as they go about their daily lives.

INTRODUCTION

Since 2015, global tuberculosis incidence has fallen by 2% per year for a cumulative reduction of 11%. Although the burden of disease has been reduced, tuberculosis remains a major cause of global morbidity and mortality, especially in sub-Saharan Africa where about 23% of global tuberculosis cases occur. New diagnostic tests and new treatment regimens for drug-susceptible and drug-resistant TB disease will improve outcomes of clinical disease and simplify management, but they may not address the persistence of tuberculosis because they have little effect on reducing transmission of M. tuberculosis before diagnosis.

The driving force behind tuberculosis persistence is the transmission of M. tuberculosis from undetected infectious cases to susceptible contacts. Although transmission is intense in households of index cases, these households account for only about 15% of tuberculous infection in a community¹. The remaining transmission occurs outside the households through contacts networks that are only partially defined. These networks comprise a wide variety of settings, such as households², health care facilities³, congregate living⁴, among others^{5, 6}, but it is unknown how each setting contributes to the persistence of infection. Although contact investigation enhances case detection and reduces case fatality, its effect on reducing prevalence of disease is minimal⁷. Meaningful steps to eliminate tuberculosis will depend on new interventions that engage the community and interrupt M. tuberculosis transmission. These interventions will require a more complete understanding of where and when M. tuberculosis is transmitted in the community.

Current metrics for tuberculosis transmission include tuberculosis notifications, disease mortality, and prevalence surveys. While these metrics may capture trends in remote and recent transmission, the timing and location of new infections are not measured. In more advanced settings, recent transmission may be estimated in different settings using whole genome sequences of recovered M. tuberculosis strains⁸, but this method provides only partial information about transmission because it does not measure incident infections nor does it include infectious cases who do not present to health care facilities for diagnosis.

The annual risk of tuberculous infection uses age-specific prevalence of infection to approximate incident infection rates⁹, but this approach is limited by sampling methods and operating characteristics of the diagnostics tests used to measure tuberculosis infection, such as the tuberculin skin test and even interferon-gamma release assays¹⁰. The shortcomings of these approaches can be mitigated by measuring incident infection directly within community cohorts¹¹. To this end, we conducted a prospective study in a closed cohort of susceptible adults to estimate the incidence rate of tuberculous infection in Kampala, Uganda, an African city with endemic tuberculosis.

METHODS

Between June 2014 and June 2016, we performed a prospective cohort study among residents of Lubaga division of Kampala, Uganda. The details of the methods are published elsewhere¹². Briefly, adults from age 17 years who had a tuberculin skin test (TST) less than 5mm at the time of screening and no signs or symptoms of tuberculosis were evaluated at one year for incident infection defined as TST conversion. Because of delays in recruitment, the duration of quarterly follow-up was extended to two years among participants who had not converted by one year.

Demographic, social, and epidemiologic characteristics were obtained through structured interviews at quarterly intervals. TST was performed by intradermal injection of 5 tuberculin units of purified protein derivative (Tubersol, Sanofi Pasteur) on the left forearm of participants. Two field workers independently read the diameter of induration using digital calipers within 48 to 72 hours. HIV testing was performed at baseline and one year following the Uganda Ministry of Health National HIV testing algorithms.

In HIV-seronegative participants, TST conversion was defined as a TST reading of 10 millimeters or greater at twelve months with an increase in induration of at least six millimeters from baseline¹³. In HIV-seropositive participants, TST conversion was defined as a TST reading of five millimeters or greater with an increase in induration of at least three millimeters¹⁴. Participants who did not meet the above conditions were classified as non-converters.

Characteristics of the cohort were summarized with proportions and measures of central tendency. We estimated incidence density¹⁵ for TST conversion over the study period. Incidence density of new tuberculosis infection was determined as the number of TST conversions divided by total person-time of participants with complete outcome information at follow-up. We used incidence density to calculate the cumulative proportion of conversion at 12 months of follow-up¹⁵. We performed a similar analysis to estimate the incidence density of HIV infection.

Because of the potential for boosting of the TST from repeated testing, lack of specificity of the TST, and censoring of participants, we performed a sensitivity analysis to evaluate the potential effects of each of these factors individually and as a group. To assess the effect of boosting, we adjusted the cumulative incidence for a 2% false-positive rate¹⁶. To adjust for the lower specificity of TST, we adjusted for concordance with interferon-gamma release assays as reported in the literature¹⁷. Because some participants did not complete follow-up with a close-out TST, we performed an analysis assuming that none of these censored participants became infected.

A Poisson regression model with offset of observation time was used to estimate rate ratio for infection in our cohort. We performed univariate analysis to estimate crude rate ratios for association between TST conversion and a set of explanatory variables. We performed a multivariable analysis that included potential confounders and factors associated with TST conversion in a univariate analysis (p-value ≤ 0.10). Under the criterion of model independence, the quasi-likelihood was used to compare models and determine the set of risk factors that showed the best model fit. To evaluate the effect of potential selection bias from losses to follow-up, we performed a sensitivity analysis using multiple imputation by chained equation¹⁸. We imputed missing values in covariates and TST outcome for all 1681 participants by (1) including all variables in an imputation model to generate 50 imputed datasets by a conditional specification algorithm; (2) estimating rate ratios using a modified Poisson regression for each imputed dataset; and (3) combining multiple rate ratios from the 50 datasets using Robin’s formula¹⁹ to obtain pooled estimates. R version 3.6.1 was used for analysis.

RESULTS

We screened 8552 persons to enroll 1681 participants. Of these participants, 406 (24%) participants were lost-to-follow up or refused TST at follow-up (Figure S1), leaving 1275 available with complete follow-up information. Except for history of exposure to a TB case, the distribution of characteristics among these participants did not differ from those with complete outcome information (Table S1; Table S2). Among the 1275 participants, the median age was 24 years (IQR 20 – 29), 41% were male, and 5% were HIV-infected at enrollment (Table 1). The proportion of participants who had been in known contact with a tuberculosis case either over one year before enrollment or during follow-up or both was 18% (Table 4).

Table 1.

Characteristics of 1275 participants with available TST results at one year follow-up

Characteristics	Frequency (%)
Age in years, median (IQR)	24 (20, 29)
Sex
Female	747 (58.6)
Male	528 (41.4)
Monthly income, Ushs
99,999 or less than	453 (35.6)
100,000 – 199,999	383 (30.1)
200,000 or more	435 (34.2)
Religion
Roman Catholic	518 (40.7)
Others	755 (59.3)
Marital status
Married	707 (55.5)
Never married	568 (44.5)
Education level
None or Primary	406 (31.9)
Secondary/Post-secondary	866 (68.1)
BCG vaccination¥
No/Unknown	99 (7.8)
Probably Yes	147 (11.5)
Yes	1029 (80.7)
Smoking
Never smoked	1219 (96.1)
Former smoker	24 (1.9)
Current smoker	25 (2.0)
Alcohol use
Non-users	989 (78.7)
Light users	192 (15.3)
Heavy users	75 (6.0)
Type of alcohol drunk*
Beer/wine	243 (19.1)
Spirits/crude/waragi	103 (8.1)
Local brew	64 (5)
Non-users	989 (77.6)
HIV test result at baseline
Negative	1205 (94.5)
Positive	61 (4.8)
Refused	9 (0.7)

^*Total exceed 1275 due to multiple responses

^¥Yes includes confirmation by immunization card, confirmation from verbal report and presence of BCG scar, or BCG scar only; Probably Yes includes confirmation by verbal report only

Table 4.

Crude rate ratios of developing tuberculous infection according to history of exposure to a tuberculosis case before or during enrollment

Variables	Before enrollment	During follow up	Follow-up time in person-year	No. incident tuberculosis infection	Crude RR (95% CI)
Ever known anyone with TB
	No	-	1147	142	Ref
	Yes	-	301	50	1.37 (1.02, 1.84)
Ever lived in a home with TB patient
	No	-	1313	174	Ref
	Yes	-	135	18	1.06 (0.68, 1.65)
Have been in contact with TB patient
	No	No	1182	150	Ref
	No	Yes	147	18	1.02 (0.65, 1.61)
	Yes	No	70	12	1.35 (0.8, 2.3)
	Yes	Yes	50	12	1.92 (1.16, 3.18)
Live at the same house with TB patient
	No	No	1356	179	Ref
	No	Yes	46	4	0.69 (0.27, 1.77)
	Yes	No	30	6	1.41 (0.69, 2.91)
	Yes	Yes	16	3	1.52 (0.56, 4.14)
Sleep in the same bedroom with TB patient
	No	No	1410	183	Ref
	No	Yes	19	2	0.84 (0.23, 3.08)
	Yes	No	13	5	2.79 (1.41, 5.52)
	Yes	Yes	6	2	2.68 (0.91, 7.9)
Spent more than 4 hours at a time with TB patient
	No	No	1304	170	Ref
	No	Yes	84	11	1.05 (0.6, 1.84)
	Yes	No	43	8	1.37 (0.73, 2.58)
	Yes	Yes	17	3	1.43 (0.52, 3.94)
Knew contact with TB receiving treatment
	No	No	1175	157	Ref
	No	Yes	86	5	0.47 (0.2, 1.1)
	Yes	No	132	16	0.91 (0.56, 1.46)
	Yes	Yes	54	14	1.95 (1.23, 3.09)
Took care TB patient
	No	No	1326	174	Ref
	No	Yes	42	5	0.98 (0.43, 2.22)
	Yes	No	63	10	1.18 (0.66, 2.11)
	Yes	Yes	17	3	1.42 (0.52, 3.91)
TB contact coughing while being with you
	-	No	1270	166	Ref
	-	Yes	178	26	1.17 (0.81, 1.71)

Of these 1275 participants, 194 converted the TST over the study period of whom 158 converted during the first year. The overall incidence density of TST conversion was 13.2 conversions per 100 person-year (Table 2), which corresponds to an annual cumulative incidence of tuberculous infection of 12.4% (95% CI: 11.0%, 14.0%). The incidence density of new HIV infection was 1.6 infections per 100 person-year (Table 2), which corresponds to an annual cumulative incidence of HIV infection of 3.2% (95% CI: 1.1, 5.4).

Table 2.

Incidence density and cumulative incidence of tuberculous infection and HIV infection over study period

	Baseline N (%)	Follow up N (%)	Person-year of observation	N†	n††	Incidence Density per 100 person-year (95% CI)	Follow up days, median (IQR)
TST result
Positive	0 (0)	194 (11.5)	1468	1275	194	13.2 (11.6, 15.1)	379 (366, 431)
Negative	1681 (100)	1081 (64.3)
Refused	0 (0)	406 (24.2)
HIV test result
Yes	79 (4.7)	101 (6.0)	1366	1185	22	1.6 (1.1, 2.4)	379 (367, 431)
No	1557 (92.6)	1163 (69.2)
Refused	45 (2.7)	417 (24.8)

^†No. of susceptible participants for developing HIV or latent tuberculosis infection

^††No. of incident HIV infection or latent tuberculosis infection

The cumulative incidence of tuberculous infection over a one-year period varied by demographic characteristics (Table 3). Men experienced a higher incidence than women (14.9% versus 10.5%). The cumulative incidence of infection increased with age, ranging from 7.5% among participants 17 to 21 years of age to 16.0% among participants 30 years or older. The highest cumulative incidence proportions were found among smokers (22.1%), HIV seropositive participants (19.3%), and heavy alcohol users (18.4%).

Table 3.

Cumulative incidence of tuberculous infection at one-year stratified by participant characteristics

Variables	Follow-up time in person-year	No. of tuberculosis infection	Incident density per 100 person-year	Annual risk of infection
Age category (tertiles)
17 – 21	461	36	7.8 (5.4, 10.3)	7.5 (5.3, 9.8)
22 – 29	664	98	14.8 (12.1, 17.5)	13.7 (11.4, 16.1)
30 – 60	344	60	17.4 (13.4, 21.4)	16.0 (12.5, 19.3)
Sex
Female	854	95	11.1 (9.0, 13.2)	10.5 (8.6, 12.4)
Male	614	99	16.1 (13.2, 19.0)	14.9 (12.4, 17.3)
Monthly income, Ushs
99,999 or less than	520	54	10.4 (7.7, 13.0)	9.9 (7.5, 12.2)
100,000 – 199,999	436	61	14.0 (10.7, 17.3)	13.1 (10.1, 15.9)
200,000 or more	508	79	15.5 (12.4, 18.7)	14.4 (11.7, 17.1)
BCG vaccination
No	112	16	14.3 (7.8, 20.8)	13.3 (7.5, 18.8)
Probably Yes	171	18	10.5 (5.9, 15.1)	10.0 (5.7, 14.0)
Yes	1186	160	13.5 (11.5, 15.4)	12.6 (10.9, 14.3)
Smoking
Never smoked	1404	178	12.7 (10.9, 14.4)	11.9 (10.3, 13.4)
Former/Current smoker	56	14	25.0 (13.7, 36.3)	22.1 (12.8, 30.4)
Alcohol use
Non-users	1143	143	12.5 (10.6, 14.4)	11.8 (10.1, 13.4)
Light users	215	30	14.0 (9.3, 18.6)	13.0 (8.9, 17.0)
Heavy users	88	18	20.4 (12.0, 28.8)	18.4 (11.3, 25.0)
HIV test result at baseline
Negative	1389	179	12.9 (11.1, 14.6)	12.1 (10.5, 13.6)
Positive	70	15	21.4 (11.8, 31.0)	19.3 (11.1, 26.7)

We observed a gradient of effect on cumulative incidence according to the contact’s report of exposure to tuberculosis cases (Table 4). Participants who did not report any known, prior exposure to a tuberculosis case had the lowest incidence rate; in this group, known exposure during the follow-up period of the study did not greatly increase the incidence rate. Among participants who reported prior exposure to a tuberculosis case, however, the incidence rate substantially increased with the highest incidence among those reporting exposure during follow-up (Figure S1).

In multivariable analysis (Table 5), incident tuberculous infection was associated with the middle and highest tertiles of age (22 – 29 years aRR 1.89; 30 – 60 years aRR 2.16), male sex (aRR 1.48), and known contact with at least one tuberculosis case before enrollment or during follow-up (aRR 1.84). The multiple imputation analysis among all 1681 participants produced similar results to the analysis among participants with complete outcome and other variable information.

Table 5.

Univariate and multivariable analysis with modified Poisson regression to identify risk factor associated with acquiring new tuberculous infection

Variables	Follow-up time in person-year	No. of incident tuberculosis infection	Crude RR (95% CI) (N=1258)	Adjusted RR (95% CI) (N=1258)	Adjusted RR (95% CI) with multiple imputation (N=1681)
Age category (tertiles)
17 – 21	455	36	Ref	Ref	Ref
22 – 29	652	97	1.88 (1.31, 2.71)	1.89 (1.31, 2.73)	1.82 (1.29, 2.56)
30 – 60	340	59	2.19 (1.48, 3.24)	2.16 (1.45, 3.22)	2.12 (1.45, 3.09)
Sex
Female	842	94	Ref	Ref	Ref
Male	606	98	1.45 (1.11, 1.89)	1.48 (1.14, 1.92)	1.45 (1.12, 1.89)
Monthly income, Ushs
99,999 or less than	517	54	Ref
100,000 – 199,999	425	60	1.35 (0.96, 1.91)
200,000 or more	502	78	1.49 (1.07, 2.06)
Smoking
Never smoked	1391	177	Ref
Former/Current smoker	56	14	1.96 (1.22, 3.16)
Alcohol use
Non-users	1136	143	Ref
Light users	212	29	1.09 (0.75, 1.58)
Heavy users	87	18	1.64 (1.05, 2.57)
HIV test result at baseline
Negative	1375	179	Ref
Positive	64	13	1.55 (0.93, 2.60)
Contact with TB case
No known contact before or after enrollment	1182	150	Ref	Ref	Ref
No contact before but contact after enrollment	147	18	0.97 (0.61, 1.53)	0.87 (0.54, 1.38)	0.85 (0.54, 1.35)
Contact before but not after enrollment	70	12	1.36 (0.79, 2.34)	1.32 (0.76, 2.30)	1.29 (0.76, 2.18)
Contact before and after enrollment	50	12	1.88 (1.11, 3.20)	1.84 (1.11, 3.06)	1.79 (1.09, 2.94)

In the sensitivity analysis of TST conversion at one year, we first evaluated the effect of boosting on the estimate of cumulative incidence by removing the expected false-positive TST results due to boosting. We assumed that boosted responses occurred in 2% of the cohort¹⁶ and 26 of the 158 conversions to boosting, leaving 132 incident infections. The adjusted cumulative incidence was reduced to 10.4% (95% CI: 8.8,12.2). When we assumed that cutaneous anergy was present in 25% of HIV-seropositive persons infected with M. tuberculosis²⁰, incident infections increased to 137, yielding a cumulative incidence of 10.7% (95% CI 9,2, 12.6). Using the positive predictive value of 64% for the TST in Kampala¹⁷, we estimated that 88 of the 137 TST converters were classified as true converters, yielding an adjusted cumulative incidence of 6.9% (95% CI: 5.6, 8.4).

DISCUSSION

In Kampala, an African city with endemic tuberculosis, we found that the incidence rate of M. tuberculosis infection was high, leading to an annual risk of infection of 13%. The incidence rate was higher in men and increased with age. Although awareness of recent or remote exposure to a person with tuberculosis was associated with higher rates of infection, the background incidence of infection among residents without these risk factors was high, suggesting that even community residents had adequate contact for infection with undetected, infectious cases of tuberculosis as they go about their daily lives. It is this scenario of susceptible contacts mixing with undetected infectious cases in unknown settings and places that maintains tuberculosis in the city.

The level of active disease and latent infection in Kampala has remained high over the past two decades. From community-based surveys of disease in two divisions in the city, the prevalence of tuberculosis was 746 cases per 100,000 in 2009²¹ and 940 cases per 100,000 in 2019²². The prevalence in the later study may be higher because more sensitive diagnostic tests were available. From two community surveys of latent tuberculosis infection among residents 15 years or older, the prevalence of infection (e.g., TST ≥ 10 mm) was 33% in Kawempe division in 2004²³ and 41% in the adjacent division of Lubaga in 2014²⁴. The rate of infection we observed during the corresponding period may indicate the level of M. tuberculosis transmission needed to sustain such high levels of infection and disease.

In this community cohort, the incidence of infection increased with age for both men and women. This finding contrasts with the risk of infection found in households where infection rates do not vary by age or sex²³. The difference in measured risk may be explained by comparing the nature of exposure between households and the community²⁵. In index case households, contacts are repeatedly exposed to the index case until diagnosis is made and treatment given. This level of exposure over time appears similar across sex and categories of age. The measure of risk is conditioned on household exposure, so the secondary attack rate is calculated. In the community, however, exposure is not measured or known, so the measure of risk is not conditioned on exposure; in this circumstance, the attack rate is estimated, as we did in this cohort.

While it is a longstanding tenet in the field that transmission of M. tuberculosis requires prolonged, close contact with an infectious case who coughs²⁶, we now know that transmission may occur from cases with pauci-bacillary disease²⁷, without cough²⁸, through casual contact in the community²⁹, and via aerosols produced through tidal respiration³⁰. These studies illustrate the ways in which M. tuberculosis may be transmitted undetected in the community. For the most part, exposures to infectious cases without cough may go unnoticed or not be remembered by the contacts. Thus, we must infer risk of exposure and transmission from the demographic and epidemiologic characteristics of TST converters. In this study, reported contact with a tuberculosis case increased the annual risk of infection by 84% above the already high background rate of infection.

In most parts of the world, reported cases of tuberculosis in men outnumber cases in women. Although this pattern may be due to differences in access to health care, case-finding surveys in endemic regions which mitigate referral bias still find male predominance of tuberculosis³¹. This pattern of tuberculosis may emerge because men are more often infected with M. tuberculosis or because men are more susceptible to progression of infection. As reported previously, the male predominance may be partially explained by the higher rate of infection among men compared with women due to patterns of assortativity by age and sex in social networks³².

The current study used TST conversion as the indicator of new infection with M. tuberculosis. Although the operating characteristics of the TST are known in Africa, it may produce false-positive results because of prior BCG vaccination, infection with environmental mycobacteria, and boosting¹³ or false-negative results because of immunosuppression from HIV infection³³. To delineate the effect of these factors on our estimate of incidence, we performed a sensitivity analysis that produced a lower boundary estimate for the annual risk of infection of 6.9%. It is possible that our incidence rates were overestimated because exposure was more often reported among participants who completed follow-up compared with those who were censored.

Another important limitation is that the current design does not indicate where the transmission occurs within the geographic boundaries of the target population. Apart from measuring the magnitude of transmission, the study does not provide actionable information for tuberculosis control. To reduce transmission of M. tuberculosis in the community, we need to design and test community interventions that interrupt transmission. To do this, we need a better understanding of where and in what settings M. tuberculosis is transmitted³⁴, apart from the index case households. Future cohort designs must be expanded to include geographic and mobility networks to assess how tuberculosis cases and susceptible community contacts mix. When the information from these networks is combined with the phylogenetic information of transmitted strains of M. tuberculosis, it will be possible to identify transmission networks and ‘hotspots’ of transmission. This type of information will help take the next step in building a foundation for community interventions to reduce transmission and prevent future cases.