Design

This study uses data on hourly ambient PM2.5 concentration levels in micrograms per cubic meter (μg/m3) made using a Beta Attenuation Monitor (BAM-1022). The associated mortality burden is estimated using the World Health Organization’s AirQ+ tool [7].

PM2.5 monitoringStudy site

This study was conducted in Kampala, the capital city of Uganda with an estimated population of 1,709,900 [8]. It covers an area of 189.3 square kilometers and has an estimated population density of 9352 per square kilometer. There are two distinct seasons characterized by high and low rainfalls. The dry seasons are June to August and December to February while the wet seasons are March to May and September to November. The PM2.5 sampling site was located on the roof of the Makerere University School of Public Health (MakSPH) within Mulago National Referral Hospital complex in Kampala city.

Data collection and instrument maintenance

The PM2.5 data utilized came from the continuous real-time measurement of PM2.5 concentration by a BAM-1022. The BAM-1022 has an internal working environment that maintains a steady Flow Rate (FR) (at 15.8 < FR < 17.5 liters per minute) and a range of meteorological conditions: Relative Humidity (RH) (10% < RH < 99%), Temperature (T) (1 < T < 40 °C), and Barometric Pressure (BP) (650 < BP < 670 mm Hg). After exploring the variability of PM2.5 concentrations, data were averaged over 1-h intervals, yielding 24 data points daily. For quality control, the instrument provides error messages so that problems could be addressed when needed and the BAM 1022 was carefully maintained.

Our team regularly performs background PM2.5 concentration calibration offset in ug/m3. The background value is a correction offset for the concentration data collected by the BAM-1022. The background PM2.5 concentration calibration offset is factory-calibrated but also during the BAM 1022 set up this value is verified and adjusted if necessary) prior to first use and then every 6 months using the BX-302 Zero Filter Calibration Kit. This test corrects the background PM2.5 concentration calibration offset (BKGD) value to compensate for minor variations caused by local conditions such as grounding and shelter characteristics. To measure PM2.5, a vacuum pump draws airborne particulate matter of 10 microns and less through a size-selective inlet (the PM10 inlet head), down into a PM2.5 very sharp cut cyclone where separation of airborne particulate matter of size greater than 2.5 microns takes place, allowing airborne PM of size ≤2.5 micrograms to move down the inlet tube and deposit on a filter tape located between the beta source and detector. The accumulation of mass onto the filter tape increasingly attenuates beta ray transmission through the media. This study used data for the period 1st January 2018 to 31 December 2021 (Fig. 1).

Fig. 1: BAM 1022 reference instrument.

a Investigators' site visit to Makerere University School of Public Health rooftop. b Ugandan Principal Investigator with the Air Quality Management District (AQMD) officer performing a quality assurance check.

Mortality data source

We obtained mortality data from the Uganda Bureau of Statistics (UBOS) as well as the District Health Information Software 2 (DHIS2), Division of Health Information Management Ministry of Health, Uganda.

Data management and quality assurance

Data were downloaded weekly from the BAM-1022 and evaluated according to the Hub’s standard operating procedure. For analysis, we used data for days with at least 18 of 24 h available. The missing data primarily resulted from power outages and periods of routine maintenance. Missing data were detected in the system software records by a signal for a sampling error resulting from data outside of the acceptable ranges for the following selected parameters: flow rate, concentration, temperature, barometric pressure, relative humidity, and sampling time interruption.

Among the techniques proposed in the literature for replacing missing values, we selected substitution of the mean, an imputation technique often used for air pollution data [9]. We adapted the “before-after-mean” method, which replaces all missing values with the mean of one datum before the missing value and one datum after the missing value If at least one of the before-and-after data points were not available, we moved to a 2-day window to complete the data. As a result of this data cleaning process, we recovered 120 days (8.2%) of the expected data.

Data analysis Exposure PM2.5 data analysis

The averaged continuous ambient PM2.5 measurements were used for statistical analysis. The R statistical software (R 3.6.2; https://www.r-project.org/) was used to conduct descriptive analyses. The 1-h BAM 1022 data were aggregated to create the daily averaged data sets for PM2.5. Time series line graphs were used to explore daily patterns and seasonality.

Attributable death estimation

We used the WHO AirQ+ tool to calculate the deaths attributable to PM2.5 [7]. We employed averaged concentrations over the 4 years from the BAM-1022 for the calculation of attributable mortality. The total population of Kampala for the 4 years from 2018 to 2021 was provided by the Uganda Bureau of Statistics (UBOS).

Input variables

We obtained the required input data for the AirQ+ tool kit from a variety of sources. These variables included Kampala area (189.3 square kilometers of land and water), population density (9352 per square Kilometer), and Kampala GPS coordinates (0°18′ 49″ N, 32° 34′ 52″ E). The proportion of the population above 30 years was ascertained, and the number of accidental deaths subtracted before arriving at the annual mortality data for use in the AirQ+ tool kit. Deaths resulting from accidents (external causes) are excluded because they are not likely to be related to air pollution. The annual mortality data was captured from the UBOS and the rest of data for all health facilities in Kampala city was obtained from the Ministry of Health of Uganda, Health Information Management Division. Annual PM2.5 was computed based on the daily PM2.5 captured using the BAM 1022 (Table 1).

Table 1 Input variables for health impact assessment of air pollution.

The four WHO annual interim target options for PM2.5 (Interim target 1—35 μg/m3, interim target 2—25 μg/m3, interim target 3—15 μg/m3, and interim target 4—10 μg/m3) and the WHO annual average air quality guideline for PM2.5 (5 μg/m3) were used as cut-off reference values to estimate the excess deaths attributable to PM2.5 pollution as measured by the BAM 1022. The cut-offs used to estimate the mortality due to air pollution were adopted from the AirQ+ software tool and the interim targets from the WHO global air quality guidelines [1].