Impact of Solid Fuel Use on Household Air Pollution and Respiratory Health in Two Low‑Income Communities in Mpumalanga, South Africa

Study area

The study took place in the communities of KwaZamokuhle and eMzinoni in Mpumalanga Province, South Africa, between July 2019 and February 2020 (Figure 1). A large proportion of households use coal for cooking and heating despite having access to electricity [20]. In the study area, coal is readily available and cheap given the community’s proximity to coal mines [20].

Figure 1

Community, household and participant selection

The study originally aimed to enrol 1000 participants, 500 per site and sites selected as part of a large, non‑randomised control trial aimed at identifying the health impacts of HAP reduction interventions in low‑income communities. The towns were selected based on similar demographics, geography and proximity to industrial air pollution sources such as coal mines and power stations. The study was prematurely halted after baseline data collection phase due to the COVID‑19 pandemic. The cross‑sectional data for these two communities were applied in the analyses here.

Households living in formal houses (defined as permanent residential structures with access to municipal services) across both communities were recruited. Dwelling selection took place according to a matched pair strategy. At least 20 cluster‑matched pairs of household groupings were selected based on local criteria, e.g., proximity to a school, clinic, busy road or police station. This design allowed for analysis to be done at a household level adjusting for confounding variables and reducing variability within the data. Within each of these clusters, households were randomly selected. Depending on the consent response of the study participants and other external limiting factors (e.g., the unavailability of a household member during a survey or the onset of lockdown during COVID‑19) the sample sizes for the various surveys differed (Figure 2).

Figure 2

Household and general health survey

A household questionnaire (Supplementary Table S1) was administered to all participating households after having been field validated and piloted. Data on demography (e.g., age, education and employment status), socio‑economic status (e.g., monthly income), housing conditions (e.g., water source), fuel use patterns (e.g., main fuel used for heating and cooking) and smoking practices (e.g., whether a household member smokes) were collected. The household questionnaire was completed by the woman in the household who was most frequently tasked with cooking.

Fieldworkers from the local communities were recruited and trained to administer the questionnaires. Responses were logged using mobile tablets and mobile data capture software called RedCap [21]. Pilot study results were not included in the final dataset. A random sample (~10%) of interviews was selected for quality control re‑interviews via telephone. A selection of questions was re‑asked of participants to confirm that the original answers were true.

Indoor air pollution monitoring

Airborne PM_2.5 was sampled inside a subset (n = 113) of the households across both communities using battery‑operated Airmetrics MiniVol samplers with interchangeable impactors and with continuous TSI DustTrak monitors for 24‑hour periods in each house (DustTrak II Aerosol Monitor 8530) [22]. Monitoring took place during winter and summer between July 2019 and February 2020. The subset was necessary due to the number of available instruments and budget constraints.

Concentrations measured by the MiniVols were determined gravimetrically. A microgram scale was used to weigh the filters pre‑ and post‑exposure. PM_2.5 concentrations in the households were calculated using the change in filter weight and the volume sampled during the collection of PM. The weight of unexposed filters, as well as adjusted flow rates, was taken into consideration. Gravimetric analyses were undertaken at the laboratories of the University of KwaZulu‑Natal. The cleaning, leak checks and single‑point flow rate checks were conducted for the MiniVols as per the manufacturer’s specifications.

The DustTrak instruments were factory calibrated before being deployed in the field and then zero calibrated and flow checked as per the manufacturer’s directions. Concentrations were expressed in micrograms per cubic meter (µg/m³) per 24‑hour period on weekdays.

Ambient air quality monitoring

Ambient air quality data were sourced from government‑owned, centrally located monitoring stations in each community, and each household was assigned ambient air quality data from the station nearest to them. Hourly averages were downloaded in µg/m³ for the same period for which indoor monitoring took place. These concentrations were then averaged per day for direct comparison with the indoor measurements.

Lung function test

Participants from consenting households took part in lung function testing using the Easy One (NDD Medizintachnik AG) Spirometry device [23]. The lung function tests were administered by a qualified pulmonary technologist using the American Thoracic Society guidelines for conducting spirometry. Spirometry was performed in a sitting position without nose clips, and participants were asked to blow into the device. A clean mouthpiece was used for every individual. Indices of primary interest included forced vital capacity (FVC) and forced expiratory volume in one second (FEV1) both measured in litres. The FEV1/FVC ratio was used in the diagnosis of obstructive airways disease. The ratio represents the proportion of a person’s vital capacity. In obstructive airways disease, the FEV1 is reduced due to an obstruction of air escaping from the lungs. This could, for example, be due to inflammation in the airways from exposure to air pollution [24]. The diagnosis of obstructive airways disease is made when the FEV1/FVC ratio is less than 0.7 [25].

Allergen sensitivity test

An allergy screening test for inhalant allergens was conducted. The immunological status of participants was assessed by means of a blood IgE test (a non‑specific test for allergy to allergens). A qualified phlebotomist took blood specimens. The blood was transported using standard operating procedures for analytic laboratories from the test site to the laboratory service provider. A positive result indicated that the participant’s immune system has produced IgE antibodies in response to exposure to allergens [26]. The test is considered positive if the total IgE antibody is equal to or above 0.35 KU/L [27]. This sensitisation is the first step to developing allergies but does not automatically mean that the person will show clinical allergy symptoms [26].

A comparison of respiratory health and pollutant exposure across communities

A comparison was undertaken to assess the impact of differing levels of residential fuel use and PM exposure on respiratory health in the two communities. Data were analysed using Stata 18 [28] and were summarised descriptively using frequencies for categorical variables and medians and interquartile ranges for the continuous variables as these were non‑normally distributed. The prevalence of health outcomes was evaluated per community. Respiratory health outcomes were categorised into self‑reported and objectively measured health outcomes. Cross‑tabulations were conducted to quantitatively analyse the relationship between exposure variables and measured health outcomes. Ambient and indoor air pollution daily means were calculated and compared to the National Ambient Air Quality Standards (NAAQS) [29].

Tests for differences in air pollution levels and health outcomes between communities

The Mann–Whitney U test was used to study the differences between indoor and ambient air pollutant concentrations and the respiratory health of participants between the two communities. The chi‑square test and Fisher’s exact test were used to evaluate associations between categorical variables like fuel use for larger and smaller sample sizes, respectively. p‑values were considered statistically significant if they were less than 0.05.

Regression analysis to assess associations between exposure and health outcomes

Multiple logistic regression analyses were performed using the pooled, binary lung function and allergen sensitivity test results as health outcomes (i.e., the presence or absence of obstructive airways disease or the positive or negative allergen sensitivity result). The variable ‘town’ was used as a surrogate for exposure. Based on the results of the fuel use and air quality assessments, eMzinoni was treated as the reference category. Crude and adjusted odds ratios (OR) with 95% confidence intervals (CI) were calculated to examine the relationship between measured health outcomes and a range of independent variables linked to fuel use (as HAP proxies) as well as identified confounders. Confounding variables were selected based on previous findings as well as univariate analyses [30]. We also made sure that these variables were not highly correlated with each other and identified the best statistical model using established criteria, namely the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). In the multivariable models, we adjusted for employment status, monthly income, education and age, with eMzinoni specified as the reference site.

Sequential modelling of exposure and health outcome associations

Three models were run separately for each exposure and health outcome combination. The first was the baseline bivariate analysis, or the unadjusted model (called ‘baseline’). The second model adjusted for the living conditions of the participants that might have influenced exposure levels (e.g., how long they lived in their community and whether they had a ceiling or not) (‘living conditions’). Finally, the third model adjusted for socio‑economic factors, including employment status, education level, age and monthly income (‘socio‑economic factors’).

Justification for exclusion of PM_2.5 measurements in regression analyses

Measured indoor PM_2.5 concentrations were not used as exposure variables in the bivariate or multivariate regression analyses as one daily average PM_2.5 reading per dwelling was not considered sufficiently representative of the highly variable HAP levels to which a participant is exposed. As ambient PM_2.5 measurements represent concentrations measured at one stationary location, they were not considered an adequate exposure proxy. Fuels used for heating and cooking, as well as stove types used, were grouped into ‘non‑electric’ and ‘electric’ categories. Electric categories were treated as reference categories.

Sample description

While we enrolled 1005 participants overall, a total of 908 participants completed both the household questionnaire and general health surveys (Figure 3). Fewer participants agreed to participate in the lung function and allergen sensitivity tests. Indoor air pollution monitoring was conducted in a subsample of participants’ dwellings due to a limited number of instruments.

Figure 3

Socio‑demographics of the participants

All participants were female and over the age of 18 (Table 1). The proportion of participants who completed primary or high school was twice as high in eMzinoni compared to KwaZamokuhle. The employment rates of participants were the same in both communities. Almost a third of all participants stated that their household income was less than ZAR 2000.00/month (USD 106.00). More participants in KwaZamokuhle (n = 36) featured below the food poverty line (USD 40 per month) than in eMzinoni (n = 11). After adjusting for employment status, monthly income, education and age (with eMzinoni as the reference site), the observed differences between KwaZamokuhle and eMzinoni were found to be significantly different (Table 2).

Sources of air pollution and fuel use patterns

Despite widespread electricity access for cooking, 55% of participants in KwaZamokuhle and 25% in eMzinoni used coal for cooking. This highlights the ongoing energy insecurity and potential exposure to harmful pollutants. These fuel use patterns reveal key differences between the two communities studied that may compound environmental health risks. A third of participants in KwaZamokuhle (33%) and more households in eMzinoni (41%) reported that they have stoves that made their homes smoky (Table 2).

Coal was the most commonly used fuel to heat homes. Coal was used for cooking by more than twice as many households in KwaZamokuhle than in eMzinoni and for heating in more households in KwaZamokuhle than in eMzinoni. LPG and wood were used by between 1 and 2% of households as a main cooking fuel.

Electric and LPG stoves were used by slightly more households in eMzinoni while cast iron and welded stoves for solid fuels were used in more households in KwaZamokuhle than in eMzinoni (Table 2). A statistically significant difference was found between the communities for both cooking and heating energy sources. There was also a statistically significant difference in the types of stoves used between the two towns (Table 2). Almost all households used electricity for lighting.

Only 7.5% of individuals in the households smoked. Among these, only half smoked inside the house (n = 38 across both communities). The most reported potential source of indoor air pollution was dust followed by mould and pets.

Indoor and ambient air quality

Indoor and ambient PM_2.5 levels were more than twice as high in KwaZamokuhle than in eMzinoni (Figure 4). Indoor concentrations showed large interquartile ranges (IQR) (Figure 4). The difference between the average indoor and ambient PM_2.5 concentrations in the two towns was statistically significant.

Figure 4

Measured lung function and allergy susceptibility

Participants from KwaZamokuhle had a higher median FVC (2.99 L, IQR 0.52) compared to those from eMzinoni (2.84 L, IQR 0.98). Similarly, the median FEV was higher in KwaZamokuhle (2.59 L, IQR 0.62) than in eMzinoni (2.27 L, IQR 0.93). Medians are reported instead of means due to non‑parametric distribution of the data, with interquartile ranges used to reflect variability and reduce the influence of outliers, which is common when reporting data from respiratory diseases [31]. The FEV1/FVC ratios were statistically significantly different between the two towns with participants from eMzinoni having a lower median ratio compared to participants from KwaZamokuhle (Table 3). Participants in both communities showed a non‑age‑standardized obstructive airway disease prevalence of 9%.

The median concentration of allergen‑specific IgE antibodies detected in the blood samples of participants in eMzinoni was higher than in KwaZamokuhle, but this was not statistically significant. About one quarter of all participants had positive allergen sensitivity test results (>0.35 kU/L) suggestive of sensitisation to the specific inhalant allergens of the Phadiotop test and the potential to react upon exposure.

Analyses showed no statistically significant difference in PM_2.5 exposure levels between participants with obstructive airways disease and those without the condition (Figure 5A). The same was true for the PM_2.5 concentrations to which those were exposed who presented with positive and negative allergen sensitivity test results (Figure 5B). Nevertheless, in both communities, the median indoor concentration to which allergen‑sensitive individuals were exposed was higher than that to which non‑sensitive individuals were exposed, though only marginally so in KwaZamokuhle (Figure 5B). Only in KwaZamokuhle were the indoor PM_2.5 concentrations higher in those households in which participants who were diagnosed with obstructive airways disease lived.

Figure 5

The median values of FEV1/FVC and Phadiatop concentration were not statistically different across the different fuel uses, stove types used or the presence and absence of stove smoke from the stove in both communities (Tables S2–S5).

While direct associations between fuel use, stove type, and respiratory outcomes were not statistically significant, likely due to limited sample size or unmeasured confounding, the observed trends, particularly in heating fuel use and obstructive airway disease, indicate possible associations. Importantly, ORs > 2 for heating fuel use suggest potentially meaningful health impacts (Table 4).

Notably, the ORs for non‑electric heating fuel use were consistently above ‘2’ across models, suggesting a possible elevated risk for obstructive airway disease, even though statistical significance was not reached. Wide confidence intervals underscore uncertainty but also reflect real‑world variability that needs to be captured in larger or more targeted studies. Notably, the ORs for the other exposure variables fluctuated around ‘1’ for the ‘baseline’ and the ‘living conditions’ models. Overall, the results suggest positive associations between solid fuels used for heating and decreased respiratory health, which could compound with time.

Strengths, limitations and recommendations for further research

A key strength of this study is the rare combination of measured respiratory health and household air pollution (HAP) exposure data in these low‑income communities in Mpumalanga, which provides an important evidence base for understanding local health risks. Data were collected directly from participants using standardised protocols, ensuring internal consistency. The study also adjusted multiple regression models for relevant confounders, strengthening the validity of the effect estimates. Despite the lack of statistically significant associations, the presence of effects in the expected direction suggests that exposure‑related adverse effects may be occurring in this sample. This study also provides valuable contextual insights that can inform the design and implementation of larger, more robust investigations in similar settings.

Several factors may have contributed to the absence of statistically significant associations between the exposure variables (fuel and stove types) and lung function‑related outcomes. Exposure misclassification is an important consideration and due to limited resources, only a sub‑sample of homes was monitored for PM, and these measurements were based on single samples. Evidence from previous research [19] indicates considerable inter‑ and intra‑household variability in exposure, meaning our single‑day sampling approach may not have fully captured participants’ true exposure levels. The relatively small sample size is another limitation; the targeted sample size was not reached due to study challenges, and because the effect sizes for lung function are likely to be small in cross‑sectional studies, larger samples are typically required to detect significant effects.

Data collection was further challenged by external factors, including the COVID‑19 pandemic, which restricted the collection of measured health data. Community unrest, the absence of household members during scheduled visits, and inaccessible roads due to heavy rainfall also contributed to the reduced sample size. Owing to resource constraints, one daily average measurement per dwelling was used to determine indoor PM exposure, although multi‑day (≥48‑hour) measurements are generally preferred. Ambient PM_2‧5 concentrations were measured at centrally located monitoring stations, and future studies should incorporate multi‑day and continuous monitoring to improve exposure assessment.

Future research should employ larger sample sizes and longitudinal designs to better elucidate mechanistic associations between HAP exposure and respiratory health outcomes. This should include multi‑day and continuous exposure data, early‑life exposures and post‑bronchodilator spirometry. Further investigation into the role of allergen exposure and sensitisation, particularly in relation to increased morbidity among participants with obstructive airways disease, is also warranted.