Estimating Tuberculosis Transmission Risks in a Primary Care Clinic in South Africa: Modeling of Environmental and Clinical Data

Abstract

Background

Congregate settings, such as healthcare clinics, may play an essential role in Mycobacterium tuberculosis (Mtb) transmission. Using patient and environmental data, we studied transmission at a primary care clinic in South Africa.

Methods

We collected patient movements, cough frequency, and clinical data, and measured indoor carbon dioxide (CO₂) levels, relative humidity, and Mtb genomes in the air. We used negative binomial regression model to investigate associations.

Results

We analyzed 978 unique patients who contributed 14 795 data points. The median patient age was 33 (interquartile range [IQR], 26–41) years, and 757 (77.4%) were female. Overall, median CO₂ levels were 564 (IQR 495–646) parts per million and were highest in the morning. Median number of coughs per day was 466 (IQR, 368–503), and overall median Mtb DNA copies/μL/day was 4.2 (IQR, 1.2–9.5). We found an increased presence of Mtb DNA in the air of 32% (95% credible interval, 7%–63%) per 100 additional young adults (aged 15–29 years) and 1% (0–2%) more Mtb DNA per 10% increase of relative humidity. Estimated cumulative transmission risks for patients attending the clinic monthly for at least 1 hour range between 9% and 29%.

Conclusions

We identified young adults and relative humidity as potentially important factors for transmission risks in healthcare clinics. Our approach should be used to detect transmission and evaluate infection control interventions.

Caused by Mycobacterium tuberculosis (Mtb), tuberculosis (TB) remains a global public health problem and is one of the deadliest infectious diseases worldwide. Understanding TB transmission at primary care clinics is of particular public health importance in high TB/human immunodeficiency virus (HIV) burden settings, such as South Africa, and in places with a risk of transmission of multidrug-resistant and extensively drug-resistant Mtb in clinics [1]. Sub-Saharan Africa is one of the most heavily burdened TB regions. Mtb is transmitted by droplet aerosols generated when people infected with TB cough, sneeze, shout, speak, or breathe [2, 3]. For TB transmission to occur, an infected person must expel Mtb bacilli from their respiratory tract, and an uninfected person must inhale Mtb bacilli-containing aerosols. Although TB control measures have been in place since the beginning of the 20th century, Mtb transmission is difficult to measure. Currently, the preferred approach is to measure presumptive transmission by determining secondary cases through molecular and genomic epidemiology [4, 5]. This approach is expensive and not feasible in all settings. Therefore, new approaches to measure TB transmission are needed.

This study piloted a novel approach to estimate transmission risk based on environmental measurements and patient data at a South African primary care clinic. We measured indoor carbon dioxide (CO₂) levels, which indicate the proportion of exhaled, rebreathed air in a room [6–8]. We also captured aerosol droplets containing viable Mtb bacilli from contaminated air [6, 9–11] and measured humidity, which is associated with the survival of airborne Mtb [12]. We obtained clinical data on patient diagnoses, visit frequency from electronic medical records, and cough counts in waiting areas, and we tracked people’s movements through the primary care clinic. Combining the different data allowed us to assess risk factors for airborne Mtb transmission in a high TB/HIV burden setting [13].

METHODS

Study Design

We previously described the study design in detail [13]. We collected environmental data such as indoor CO₂ levels, relative humidity, frequency of coughs, and presence of Mtb DNA in the air, as well as patient data over 4 weeks from 25 July to 23 August 2019 at a primary care clinic in Cape Town, South Africa.

Study Setting

The primary care clinic offers both TB and HIV services and reproductive health and childhood immunization services, Monday to Friday, from 7 am to 4 pm. The clinic is situated within a large settlement of formal and semiformal housing where both TB and HIV are highly prevalent [14, 15]. We delineated 3 areas within the clinic: the registration area, the waiting room, and the TB treatment room (Figure 1). Furthermore, we defined 3 time periods: morning (7–10:30 am), midday (10:30 am–2 pm), and afternoon (2–4 pm).

Patient Data

Tracking Data

We used an anonymized movement tracking system (Xovis, Zollikofen, Switzerland) to monitor people’s movements (staff members, patients, and other visitors) throughout the clinic (Supplementary Figure 1). The resulting date- and time-stamped movement data consisted of a person’s height, their position recorded as x-y coordinates, and a unique signal for each person while in the clinic (Supplementary Table 1) [13]. If individuals went out of a sensor’s range and subsequently returned, they could contribute multiple signals. Thus, the number of captured signals is higher than the number of unique persons. While in the waiting room, close contacts were defined as other persons passing within a radius of 1 meter.

Clinical Data

We extracted clinical data from the electronic patient registry for all patients who visited the clinic during the study period. These data included the date and time of arrival for the clinic visit and when the patient passed by the registration desk and their age, sex, TB diagnostic results, and date of TB treatment start (if applicable).

Environmental Data

CO₂ Monitoring

Three CO₂ monitors (Digital CO₂ Monitor Carbon Dioxide Meter XE-2000, XEAST, Guangdong, China) covered the clinic’s most crowded spaces. The CO₂ monitors were installed in the waiting room, by the registration desk, and in the TB treatment room (Supplementary Figure 1). The monitors recorded indoor CO₂ concentrations (in parts per million [ppm]), temperature, and relative humidity at 1-minute intervals (Supplementary Table 1) [9, 10]. Monitors were regularly auto-calibrated [16].

Cough Monitoring

We installed a microphone (RØDE NT-USB; Sydney, Australia) near the clinic’s waiting room ceiling to continuously record sounds (Supplementary Figure 1). We used a cough detection algorithm based on MXNET’s open-source deep learning software framework to classify audio signals as coughing or other sounds (CoughSense, Seattle, Washington) [17]. In addition, we developed a cough counting algorithm to test for cough in the recorded coughs automatically. We trained, tested, and validated the algorithm model using multiple audio recordings obtained during the study period (Supplementary Table 1).

Bioaerosol Sampling and Molecular Testing

Air was sampled using mobile bioaerosol sampling devices (Dry Filter Unit 1000, Lockheed Martin Integrated Systems, Gaithersburg, Maryland). The number of Mtb genomes was ascertained from dried filters using highly sensitive droplet digital polymerase chain reaction (ddPCR) [11]. We placed 1 bioaerosol sampling device in the clinic’s waiting room and the other in the TB treatment room (Supplementary Figure 1). During data collection, each bioaerosol sampling device collected air through 2 filters over 2 time periods (morning and midday). Each day both devices collected air for about 3.5 hours, totaling approximately 7 hours per day (Supplementary Table 1).

Linkage of People Tracking Data With Clinical Patient Data

We applied several criteria to link the movement tracking system data with the clinical data. We included people who (1) passed by registration and (2) had a height of at least 140 cm according to the tracking data to exclude children; we included clinical visits of patients aged 15 years and older from the clinical data. We then combined the datasets using the time-stamp of when a person was recorded by the tracking system in the registration area and the time a patient was registered in the electronic patient registry. We identified 2 355 adult patients (≥15 years) whose visits were recorded in the clinic’s electronic patient registry from the clinical data. After linking with the movement tracking data, we included 978 unique adult patients, resulting in 1 135 clinical visits.

Statistical Analyses and Modeling

We used descriptive statistics for the environmental and patient data obtained in the different clinic areas. We calculated the number of individuals in the 3 clinic areas, the time spent in the waiting room, and the number of contacts an individual had during this time period, thus enabling the identification of highly frequented areas.

As previously described, we calculated the rebreathed air volume and ventilation rates from CO₂ and clinic presence [10, 13]. We summarized the coughs per minute in the waiting room over the 3 time periods [18, 19]. We described the number of Mtb genome copies present in each filter by time period and clinic area.

We used a negative binomial regression model to assess clinical and environmental factors associated with the number of Mtb genome copies measured in the waiting room air (Table 1). Using the mean for the environmental data (CO₂, and relative humidity), the total number of people present in the clinic, and the total number of coughs, we aggregated the data by the minute to the exact time period of the bioaerosol sampling devices. The model was estimated with Markov chain Monte Carlo in a Bayesian framework using Stan, a probabilistic programming language [20]. The results are unadjusted and adjusted risk ratio per unit increase with 95% credible intervals (CrIs). The model was adjusted for sex, age group (15–29, 30–44, 45–59, and ≥60 years), relative humidity, indoor CO₂, and frequency of cough (Table 1).

Finally, we calculated the risk of Mtb transmission per hour during the day and per each clinical visit as previously described [21]. In brief, we used the modified Wells–Riley formula considering the work of Rudnick Milton on non-steady-state situations to estimate the annual risk of TB transmission, taking into account the rebreathed air volume, time at risk, the infectious quanta of contagion, and the number of people occupying the confined space [6, 8]. The parameters we used to calculate the risk of transmission are given in Supplementary Table 2.

All analyses were performed in R software (version 3.6.0) [22].

Ethics Statement

The University of Cape Town Faculty of Health Sciences Human Research Ethics Committee (HREC/REF: 228/2019); the City of Cape Town (Project ID: 8139), South Africa; and the Ethics Committee of the Canton of Bern (2019-02131), Switzerland, approved the study.

RESULTS

Patient Data

Movement of Patients

The movement tracking system captured 14 795 unique data points corresponding to people in the clinic between 25 July and 23 August 2019. The median number of unique signals per day was 706 (interquartile range [IQR], 622–803). Most individuals visited the clinic in the morning when the highest density of individuals was found in the waiting room (Figure 2A). The median time spent in the waiting room was 24 minutes (IQR 23–27).

Patient Characteristics

After data linkage, we included 978 unique patients. Their median age was 33 years (IQR, 26–41), and 757 patients (77.4%) were female. Overall, 171 (17.5%) had a TB diagnosis at some time, among whom 153 (90.6%) had a clinical history of TB, and 16 (9.4%) had active pulmonary TB and were potentially infectious at the time of clinic visit (Supplementary Table 3). The density of potentially infectious TB patients and all other people was highest in the waiting room (Figure 1). These potentially infectious TB patients were more likely HIV-positive men who had 3 or more visits during the 4 weeks (Supplementary Table 3).

Time in the Waiting Room

The median time a patient spent in the waiting room was 41 minutes (IQR, 17–85), with a median of 62 (IQR, 16–173) close contacts (within a radius of 1 meter). There were no significant differences between potentially infectious TB patients and all other patients in the time spent in the waiting room (41 vs 43) or in the number of contacts (67 vs 66).

Coughing

The median number of coughs per day in the waiting room was 466 (IQR, 368–503). The total number of coughs was higher at midday than in the morning (495 vs 421; Table 2). The median length of coughs was 0.67 seconds (IQR, 0.47–0.91).

Environmental Data

CO₂ Levels

The median CO₂ level in the clinic was 564 ppm (IQR, 495–646). It was higher in the morning than at midday and in the afternoon (639 vs 568.7 vs 477 ppm). We measured the highest CO₂ levels in the waiting room (Table 2, Figure 2B). The share of time people experienced CO₂ levels ≥1000 ppm of the opening hours was 4.7%.

Rebreathed Air Volume

The overall median rebreathed air volume was 46.5 L/day (IQR, 22.7–74.8), and it decreased over the day (Table 2, Figure 2C). The rebreathed air volume was highest in the waiting room compared to the registration area and TB treatment room (68.1 vs 42.3 vs 9.5 L/day). The ventilation rate in the waiting room was at 12.2 L/hour (recommended ventilation rate: 6.0 L/hour [23]).

Relative Humidity

The overall median relative humidity was 60.6% (IQR, 53.6%–65.8%). It was higher in the morning compared to midday and afternoon (66.2% vs 58.9% vs 54.1%). The relative humidity was highest in the TB treatment room followed by the registration area and the waiting room (63.6% vs 60.9% vs 57.3%) (Table 2, Figure 2D).

Presence of Mtb DNA Copies/μL in the Air

The overall median number of Mtb DNA copies/μL per day was 4.2 (IQR, 1.2–9.5). The median Mtb DNA copies/μL throughout the day was slightly higher in the waiting room than in the TB treatment room (Table 2, Supplementary Figure 2A), and higher in the afternoon than in the morning.

Risk Factors for Potential Transmission

In the univariate analysis, we found an increased presence of Mtb DNA copies in the air of 15% (95% CrI, 3%–32%) per 100 incremental young adults (aged 15–29 years) visiting the clinic. No other variables were associated with an increase presence of Mtb DNA copies in the air (Table 1). In the multivariate analysis, we found an increased presence of Mtb DNA copies in the air of 32% (95% CrI, 7%–63%) per 100 incremental young adults (aged 15–29 years) visiting the clinic. For a 5% incremental increase of relative humidity, 1% (95% CrI, 0–2%) more Mtb DNA copies were in the air (Table 1). Figure 3 shows the standardized risk ratio per 1 standard deviation with the 95% CrI.

Risk of Infection

We modeled different scenarios using the observed TB prevalence at the clinic and the estimated TB prevalence of 737 per 100 000 people for South Africa with varying infectious quanta [24] (5.5 and 8.2 infectious quanta per hour; Supplementary Table 2). The observed TB prevalence at the clinic suggested that the risk of Mtb transmission during the day was about 3% per hour using 5.5 infectious quanta per hour. It was about 6% per hour using 8.2 infectious quanta per hour (Figure 4A). The risk of infection was lower when using TB prevalence estimates by World Health Organization (WHO) (Figure 4B).

To put this in perspective, a patient coming each month to the clinic for 1 hour (12 visits per year) would have a cumulative risk of Mtb transmission ranging from 9% to 29% depending on the scenario (Figure 5). The cumulative risk was higher for observed TB prevalence at the clinic compared to the TB prevalence estimated by WHO. In an extreme scenario assuming a weekly visit to the clinic of 1 hour (52 visits per year), a patient would have a cumulative risk ranging from 33% to 78%, depending on the scenario.

DISCUSSION

At this South African primary care clinic, an increased risk of Mtb transmission was associated with the presence of young adults and higher room humidity. We estimated the risk of transmission during a clinic visit of 1 hour to be 3%–6%, increasing to 9%–29% for patients making regular monthly visits. Our study suggests that multiple environmental measures and clinical data can be used to assess indoor ventilation quality and evaluate airborne disease transmission control measures in primary care and similar settings.

Our study observed more copies of Mtb DNA in the air when young adults visited the clinic than when clinic visitors were older. Several factors might account for this. Behavioral and social contact patterns differ by age, and they might play a role in the risk of Mtb transmission [25, 26]. Young index cases (<40 years) have been shown to have more close contacts and contacts with all age groups than older index cases who have fewer contacts (and mainly within their own age group) [26]. And as adolescents and young adults transition from child to adult health services, they face specific age-related challenges accessing appropriate healthcare [27, 28]. These challenges might result in delayed HIV or TB diagnoses and treatments. A study from Cape Town, South Africa, showed that TB notification was highest among young adults. Among those aged 25–45 years, 63% were HIV-associated TB patients. The study also showed that TB notification rates among HIV-negative people peaked between age 20 and 24 years and a second peak between age 45 and 54 years [29]. We observed that increasing relative humidity was associated with increased copies of Mtb DNA in the air; only a few other studies have investigated this relationship. Relative humidity was shown to play an important role in the presence of Mtb genome copies in the air [30], and our finding is also in line with results from a more recent study which showed that Mtb DNA copies were more likely to be found in health facilities when the relative humidity was >65% [12].

Studies of different settings have reported that healthcare clinics may be drivers of Mtb transmission [31–35]. In low- and middle-income countries, resource-constrained care clinics are often crowded with people sitting close together on benches or standing in passageways. In these clinics, the waiting times are often long and the ventilation is poor. These kinds of conditions favor Mtb transmission [9, 36, 37]. Because of these conditions, exposure to Mtb might be prolonged. People with undiagnosed TB or delayed TB diagnoses pose risks of Mtb transmission to other individuals at the clinic. In addition, those diagnosed with TB who continue to receive care at a clinic may pose a risk to uninfected people and reinfection in people with a Mtb infection [36]. Furthermore, HIV coinfection plays a major role as disease progression is faster in HIV-positive compared to HIV-negative individuals [31, 32]. Therefore, it is important to screen people regularly for TB symptoms. Infection control measures are needed, such as improved ventilation and, for presumptive TB cases or anyone who is coughing, wearing masks. Because of the coronavirus disease 2019 (COVID-19) pandemic, wearing masks is likely an easy and familiar intervention to implement. Finally, detection of Mtb DNA by ddPCR has been shown to be more sensitive than detection by aerosol using traditional culture techniques [11].

High indoor CO₂ levels (>1000 ppm) are indicators of poor ventilation. We found CO₂ levels above 1000 ppm, mainly in the morning in the waiting room area of the clinic. Levels in the TB treatment room were kept lower through measures to minimize occupancy and keeping the doors and windows open to allow ventilation. Since we know that crowded waiting rooms are the most likely infectious place, we focused on this room as well as the TB treatment room where presumptive TB patients are screened, diagnosed, and treated [38]. Previous studies have measured CO₂ levels at different locations and combined these environmental data with social interaction data to model the risk of Mtb transmission [9, 10, 21]. The highest annual risk for Mtb transmission in another southern African setting was found in prisons, with descending lower risks for persons in schools, riding public transport, and social halls [21]. These findings complement other studies of high-burden settings, which found that only a small proportion of Mtb transmission occurs between household members [39–41]. Using the observed prevalence at the clinic, we found that the risk for Mtb transmission during the day was 3%–5% per hour. A modeling study showed that the annual risk of Mtb infection in the waiting room at a clinic with closed windows and doors ranged from 23% to 34% for chronic patients with monthly visits and from 2.2% to 3.4% per patient visit [38]. Moreover, they showed that with good ventilation, the risk of Mtb infection was reduced 50-fold.

The mathematical models showed that the duration and frequency of clinic visits increased the risk of Mtb transmission. However, this could be addressed effectively by relatively simple infection control interventions: improved ventilation through opening windows and decreased room presence, which resulted in very low rebreathed air volume for the room. In settings where airborne transmission is possible, both Mtb bacilli and the severe acute respiratory syndrome coronavirus 2 are transmissible via aerosols [42–45]. In the COVID-19 pandemic, primary care clinics have implemented infection control measures such as increased hand hygiene and physical distancing, and all attendees and clinical staff members are wearing face masks. These infection control measures would likely also decrease the risk of Mtb infection and other airborne-transmitted diseases at healthcare clinics.

The collection of environmental data had several limitations. The video sensor system assigned a new ID whenever a seated person stood up. Therefore we had challenges in tracking people, and we cannot exclude incorrect assignments in these cases. Furthermore, the bioaerosol sampling devices collected data over about 3.5 hours, whereas the other data were collected by the minute. By aggregating these data, we lost some information, which may explain why we did not find an association between Mtb counts and CO₂ levels. The highly sensitive ddPCR assay we applied detects Mtb genome DNA but does not distinguish between viable, Mtb bacilli causing infections and dead or noninfectious bacilli and DNA fragments. Moreover, the assay could conceivably be detecting DNA fragments present in the clinic over a long time, and efforts in our laboratory are underway to develop improved analysis and assay approaches that can address this. These caveats notwithstanding, we used a novel and rapid system to study transmission, which goes beyond traditional methods such as molecular genotyping. However, we did not measure actual transmission events, but rather estimated the risk of transmission events using a range of clinical and environmental data, including detection of Mtb DNA in the air.

Our approach to assessing Mtb transmission risks using various environmental and clinical data is novel. It identified young adults and relative humidity as potentially important factors in TB transmission in these settings. A global study using the WHO TB notification database that showed about 17% of all new TB cases were among people aged 10–24 years [46]. Therefore, TB research and public health interventions should have increased focus on young adult health [29, 46, 47]. However, we should not only understand the drivers of transmission, but also evaluate interventions [48]. Our multiple measures approach can be used in healthcare clinics and other congregate settings to evaluate interventions to halt transmission, including the evaluation of infection control measures such as improved room ventilation, increased hand hygiene, or wearing of masks.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.

Notes

Author contributions. K. Z., C. M., R. W., M. E., and L. F. wrote the concept. K. Z. and L. F. wrote the first draft of the paper, which was reviewed by all authors and revised based on the comments received by all coauthors. K. Z., C. M., K. M., and R. W. coordinated data collection. C. M., A. K., K. M., D. F. W., and R. W. were involved in laboratory work, and they were involved in extracting the clinical data from the electronic registry. S. B. did the medical informatics and cough extraction using artificial intelligence (AI). J. R. and K. Z. completed the statistical analyses. All authors approved the final version of the manuscript.

Acknowledgments. We thank Mbali Mohlamonyane for setting up the bioaerosol sample collection system, and Zeenat Hoosen and Ronnett Seldon for developing the laboratory protocols. We are also grateful to the clinical manager and the staff members at the primary care clinic. Finally, we would like to thank the city of Cape Town, South Africa, for the use of one of their clinic facilities and their support. We thank Kristin Marie Bivens and Christopher Ritter for editorial assistance.

Disclaimer. Our research findings and recommendations do not represent an official view of the city of Cape Town.

Financial support. The rollout of this study was supported by the Swiss National Science Foundation (grant number CRSK-3_190781). K. Z. and L. F. are supported by the National Institute of Allergy and Infectious Diseases, National Institutes of Health (grant number 5U01-AI069924-05), and M. E. is supported by special project funding from the Swiss National Science Foundation (grant number 174281).

Potential conflicts of interest. All authors: No reported conflicts of interest.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

REFERENCES