Thermal comfort, ventilation and mosquito-borne disease risk in Zanzibar hospitals

2.1 STUDY DESIGN

The study employed a longitudinal observational design to assess environmental conditions and thermal comfort in four district hospitals in Zanzibar over a 48-day period. The research integrated continuous indoor climate monitoring with analyses of door-opening events, solar radiation simulations and thermal imaging.

2.2 STUDY SITE

Zanzibar is composed of two islands, Unguja and Pemba, which are situated in the south-western Indian Ocean off the east coast of Tanzania. The study was conducted at four district hospitals: Kivunge, Mwera Pongwe, Jitimai and Chumbuni, located on Unguja island (see Figure 2).

Figure 2

2.3 CLIMATE

Unguja island has a tropical humid climate. Seasonal variation is driven by two distinct rainy seasons, the long rains (masika, March–May) (Kijazi & Reason 2005) and the short rains (vuli, October–December) (Osima & Kijazi 2020). Average monthly temperature data, collected from Zanzibar weather station located at Zanzibar airport, show lows ranging from 20.5 to 24.1°C (June to August) and highs of between 31.6 and 33.0°C (December to February) (Adam et al. 2020).

2.4 DESCRIPTION OF STUDY HOSPITAL SITES

The four selected hospitals, built between 2022 and 2023, share the same design with minor differences in spatial configuration. Chumbuni and Jitimai District Hospitals are located in the more densely populated Urban West Region, close to Zanzibar’s urban centre. Kivunge District Hospital is situated in a rural town in the North Region, on a site shared with older hospital buildings and surrounded by some housing and institutional buildings. Mwera Pongwe District Hospital, in the South Region, is the most isolated of the four study hospitals, set largely within farmland. The main construction consists of a concrete frame with concrete foundation and floor, ceiling slabs and concrete block walls. The study hospitals are arranged over two storeys, with central waiting areas off the main entrance that lead to double-stacked corridors, from which wards, offices and clinical spaces are accessed (see Figures 3 and 4). External windows in the wards are multipanel aluminium sliding with a sliding fly-mesh panel, allowing half the window area to be fully open or screened. The hospitals are primarily passively cooled, assisted by ceiling fans located in the centre of spaces, with air conditioning units installed in some clinical and office areas.

Figure 3

Figure 4

Three hospital space typologies were studied: inpatient wards, waiting areas and outpatient clinical and office spaces. Three inpatient ward subtypes were monitored: medical wards, providing general inpatient care for patients with a range of medical conditions; surgical wards, accommodating patients awaiting or recovering from surgical procedures; and isolation wards, designated for patients with communicable diseases or used as overflow capacity when no infectious patients are present.

2.5 INDOOR CLIMATE AND DOOR-OPENING DATA COLLECTION

Indoor climatic data were recorded in 25 spaces across the four study hospitals. Indoor air temperature (°C), relative humidity (%) and CO₂ concentration (ppm) were recorded in each space using a single Aranet4 PRO sensor, installed at a height of 800–1300 mm above floor level and positioned away from direct sunlight. Sensors were used with factory calibration settings. Prior to installation, sensors were checked against each other to confirm comparable readings across units. Sensor locations in all hospitals can be found in the supplemental data online (file 2: sensor locations).

Door-opening was recorded using Aranet Dry Hour Meters connected to a magnetic reed switch attached to the doors. Data were logged every 10 minutes over a 48-day period from 23 April 2024 to 9 June 2024. Specifications of all sensors and measurement instruments are included in Table 2. Drawings of sensor installation can be found in supplemental data online (file 1: sensor installation).

2.6 WEATHER DATA COLLECTION

Outdoor climatic data were obtained from the Open-Meteo Historical Weather API (Zippenfenig 2023). Daily data from a government weather station located in the airport (see Figure 2 for location) were retrieved for the period 23 April to 10 June 2024. Variables included daily mean air temperature at 2 m height (°C) and precipitation (mm).

2.7 DATA MANAGEMENT

Climate and door-opening data from each sensor were transmitted every 10 minutes via radio signal to an internet-connected base station, which then uploaded the data to a secure cloud server, for download as CSV files. Prior to analysis, data files were screened for missing values, sensor dropouts and anomalous readings.

2.8 THERMAL COMFORT CALCULATION

PPD was used to assess thermal comfort. PPD predicts the percentage of occupants who will be dissatisfied with thermal conditions based on the Predicted Mean Vote (PMV) model developed by P. O. Fanger (Fanger 1970). The PMV calculation incorporates air temperature and relative humidity, along with other factors such as airspeed, clothing level and metabolic rate. For this study, an assumed internal airspeed of 0.1 m/s, a clothing level of 0.6 (equivalent to wearing trousers and a long-sleeved shirt) and a metabolic rate of 1.0 (equivalent to being seated) were used. The PPD was derived from the PMV using Equation (1). The coefficients of PMV⁴ and PMV² are 0.03353 and 0.2179, respectively.

2.9 ESTIMATING THE CO₂ LEVEL DETECTABLE BY MOSQUITOES

An estimate of 530 ppm was used as a conservative, indicative reference point above which indoor CO₂ may contribute to mosquito attractant accumulation, calculated by adding the lowest CO₂ levels detected by mosquitoes (108–133 ppm; Grant and O’Connell 2007) to average atmospheric CO₂ levels (419–421 ppm; CenCO2PIP 2023).

2.10 ESTIMATING THE CO₂ LEVEL IMPACTING HUMAN BUILDING OCCUPANTS

A literature review of low-level CO₂ exposure in indoor environments reported building-related symptoms to begin at 700 ppm (Azuma et al. 2018). Mean hourly values above this absolute threshold were plotted to identify periods during which occupants may be exposed to concentrations that could impact their health and comfort.

2.11 EXTENDED DOOR-OPENING THRESHOLD

Extended door-opening was defined as greater than 10 minutes open per hour or more than 10 opening events per hour. An hourly representation was used to allow direct comparison with mean hourly PPD and CO₂ levels presented in the heatmap figures. These thresholds were defined by the study authors to distinguish incidental door use from sustained or repeated opening patterns that may present meaningful opportunities for mosquito entry.

2.12 STATISTICAL ANALYSIS OF ENVIRONMENTAL FACTORS AND DOOR-OPENING

A custom Python script (SciPy, pandas) was used to examine associations between prolonged door-opening, PPD and CO₂ levels across three wards. A binary door-status variable was derived from door logger data (open = 1, closed = 0). Point-biserial Pearson correlation coefficients and two-sample independent Student’s t-tests (equal variances assumed, two-tailed) were used to identify broad associations between door status and each environmental parameter. Results are presented as means and p-values (5%).

2.13 VISUALISATION OF ENVIRONMENTAL AND BEHAVIOURAL DATA

A custom Python script was developed for this study to analyse and visualise environmental and behavioural data. The script utilised multiple Python libraries including pandas for data manipulation, matplotlib and seaborn for visualisation and numpy for numerical operations. Input data, including temperature, humidity and CO₂ measurements, were processed from CSV files. PMV and PPD were calculated using functions based on Fanger’s thermal comfort model, as described in ASHRAE Standard 55 (ASHRAE 2017). Hourly mean PPD values above the ASHRAE recommendations (>20%) were analysed, as well as hours with PPD values estimating the majority occupant dissatisfied (>50%). The Python scripts generated PDF visualisations comprising heatmaps of hourly PPD, CO₂ levels and door logger data, as well as daily mean temperature and CO₂ levels, and hourly mean PPD and CO₂ levels. Hourly heatmaps were developed as a novel method of visualisation to map hourly environmental conditions against known mosquito peak biting times described in Table 1. Door logger data was incorporated in applicable wards to indicate periods of extended door-opening (>10 minutes/hour or >10 opens/hour).

All Python scripts have been made open-source and are available here: https://doi.org/10.5281/zenodo.21378274.

2.14 SIMULATION AND THERMAL IMAGERY

Direct solar radiation simulations and thermal images were utilised to analyse factors influencing the thermal environment in the hospitals. A digital 3D model of Kivunge Hospital was created using Rhinoceros 3D including the surrounding context and simplified tree models. The software package Ladybug was used to estimate the number of direct sunlight hours entering the glazed openings over the study period. EnergyPlus Weather files for Zanzibar were downloaded from https://www.ladybug.tools/epwmap/. Infrared images of Chumbuni, Kivunge and Mwera Pongwe hospital interiors and exteriors were taken on 16 August 2024, after the study period, using a Flir Edge One Pro camera, and analysed using Flir One Ignite.

2.15 SEMI-STRUCTURED INTERVIEWS WITH WARD INPATIENTS

Semi-structured interviews were conducted with ward patients (n = 10) and patient relatives (n = 1) across four study hospitals, covering eight wards, between August and October 2024. Interviewees were asked to describe the thermal conditions of the ward, whether they had observed mosquitoes indoors, and what protective measures had been available to them. It should be noted that some interviews are from wards where environmental data were not collected. Interviews were conducted in Swahili, audio-recorded with the consent of all participants, transcribed, and translated into English. All personally identifying information was removed prior to analysis.

2.16 WORKSHOP: WINDOW AND DOOR-OPENING IN WARDS

As part of the co-creation activities of the MBD-Free Project, a workshop activity was conducted with patients, hospital staff and relevant specialists to explore the reasons why windows and doors are left open in hospital wards (Sloan Wood et al. 2025). Participants were presented with a physical model of a typical ward and invited to brainstorm reasons why windows and doors might remain open, given that these represent potential mosquito entry points. Reasons were recorded during the activity, and the frequency with which each reason was raised was counted to indicate relative prominence across participants.

3.1 COMBINED PERCENTAGE OF HOURS FOR ABOVE ESTABLISHED MEAN PPD AND CO₂ THRESHOLDS

Table 3 shows the percentage of mean hourly values for PPD less than 20% (ASHRAE 55 standard), PPD less than 50% (predicted majority of occupants satisfied), CO₂ levels greater than 530 ppm (estimated detectable threshold for mosquitoes) and CO₂ levels greater than 700 ppm (estimated threshold where occupants start experiencing building-related symptoms) across the four study sites. Excluding the air-conditioned ground-floor office in Jitimai Hospital (PPD < 20% for 85.6% of monitored hours), all spaces were outside ASHRAE 55 recommendations for thermal comfort (PPD < 20%) for the majority of the hours recorded. Majority occupant thermal comfort (PPD < 50%), where over half of the occupants should be thermally comfortable, was achieved for 22.4% to 100% of the monitored time. Kivunge Hospital’s surgical ward performed worst, achieving majority satisfied PPD comfort levels (PPD < 50%) for only 5.4% of the monitored time.

CO₂ concentrations varied across locations. Seven of the 25 recorded spaces experienced CO₂ levels above 530 ppm for more than 50% of the monitored period. CO₂ levels above 700 ppm were less frequent but still notable in some areas, such as Jitimai Hospital’s office (70% of the time) and Kivunge Hospital’s surgical ward (60.7% of monitored time).

Waiting areas in all four hospitals generally exhibited better thermal comfort and lower CO₂ levels than ward spaces, with 77.5% of hours achieving PPD < 50% and no hours exceeding 700 ppm CO₂. Ward spaces showed the lowest percentage of hours with PPD < 20% (2.0%) and the highest CO₂ levels above 700 ppm (13.5% of hours). Clinical and office spaces had the highest mean percentage of hours with CO₂ above 700 ppm. These results are summarised across space types in Figure 5; data from all individual hospital spaces can be found in the supplemental data online (file 3: environmental data all hospitals).

Figure 5

3.2 MEAN HOURLY TEMPERATURE AND CO₂

Figure 6 illustrates the mean daily temperature and CO₂ levels during the study period at Kivunge Hospital. Similar patterns were observed in the other hospitals; graphs for all recorded hospital spaces are available in the supplemental data online (file 3: environmental data all hospitals). All non-air-conditioned spaces exhibited similar seasonal temperature fluctuations. However, ward spaces consistently remained 1–3°C warmer than waiting areas. Ward spaces experienced greater fluctuations and peaks in CO₂ levels, frequently exceeding 530 ppm and 700 ppm. Five of 25 monitored spaces (Kivunge: GF Medical Ward; Chumbuni: FF Surgical Wards; Jitimai: Office and FF Medical Ward; Mwera Pongwe: FF Surgical Ward) experienced hours with mean CO₂ levels that exceeded the ASHRAE recommendations of 700 ppm over the outdoor level (ASHRAE 2019), based on an average atmospheric 419–421 ppm (CenCO2PIP 2023). In contrast, CO₂ levels in waiting spaces remained relatively low, typically ranging between 450–575 ppm for most of the study period. Indoor temperature in ward and waiting spaces was reduced during days with rain.

Figure 6

3.3 MEAN HOURLY AVERAGE FOR PPD AND CO₂

Figure 7 shows mean PPD and CO₂ values for each hour of the 24-hour cycle across spaces in Kivunge Hospital. All non-air-conditioned spaces show a similar daily PPD pattern, increasing during daylight hours to its highest between 13:00 and 16:00. Average hourly PPD for all spaces was above the ASHRAE 55 20% recommendation. Average hourly PPD is typically 20–25% lower in the waiting spaces compared with the ward spaces. CO₂ levels in ward spaces were relatively consistent throughout the day, with increased levels in the evening hours between 18:00 and 06:00. The air-conditioned ground-floor medical ward had fewer temperature fluctuations (Figure 6) but PPD remains higher on average than in waiting areas.

Figure 7

3.4 COMBINED MEAN PPD AT 00:00, 06:00, 12:00 AND 18:00

Figure 8 shows the combined mean PPD at 12 am, 6 am, 12 pm and 6 pm across all four hospitals for the ward, office, clinic and waiting spaces. The mean PPD was greatest in the ward spaces at all four times. The difference between the PPD in the ward and waiting spaces was greater during the night hours (18:00 to 06:00) than during the day.

Figure 8

3.5 HEATMAPS OF HOURLY MEAN PPD AND CO₂

The heatmaps of hourly mean PPD and CO₂ during the study period show recurring differences between the ward, office and clinic spaces, and waiting spaces as shown in Figure 9. In the ward spaces, the majority of hours had a PPD value above 50%, with multiple hours of PPD > 50% and CO₂ levels > 700 ppm occurring, mostly at night. These hours overlap with commonly reported peak mosquito biting hours found in the literature for Aedes (Chadee 1988, 2013; Muhammad et al. 2020), Anopheles (Sherrard-Smith et al. 2019; Sougoufara et al. 2014) and Culex species. (Chen et al. 2017; Ryan et al. 2017); see Table 1. PPD levels also reached 100% for multiple hours in the wards but not in the waiting or office and clinic spaces.

Figure 9

In the air-conditioned clinic and office spaces, PPD as low as 0% was observed in Jitimai Office. However, this space also had CO₂ levels above 700 ppm for 84.7% of the hours in the study. Other air-conditioned spaces, like Chumbuni Family Planning, had multiple hours with PPD > 50% and CO₂ levels > 700 ppm, largely in the daytime between 09:00 and 16:00, reflecting use patterns.

3.6 DOOR-OPENING BEHAVIOUR

Figure 10 shows PPD and CO₂ hourly heatmaps for ward spaces in Kivunge Hospital, where door loggers had been installed on the only entrance door to the rooms. Hours where the door was logged as being open for more than 10 minutes or opened more than 10 times are marked with an ‘X’. In the surgical, ground-floor medical, and first-floor medical wards, 192 (17.0%), 438 (38.8%) and 274 (24.3%) hours, respectively, were recorded where the door was open for more than 10 minutes. The majority (73.8%) of this door-opening occurred at night between the hours of 18:00 to 06:00.

Figure 10

Statistical analysis of door-opening behaviour across three hospital wards presented in Table 4 shows varying relationships between environmental factors and door status, where ‘open’ indicates the door was left open for more than 10 minutes in an hour or opened over 10 times. In the ground-floor medical ward, weak correlations were observed between extended door-opening and both PPD (–0.0447) and CO₂ levels (0.0318), with no statistically significant differences in these parameters between hours with extended door-opening and those without.

The first-floor medical ward demonstrated a stronger negative correlation between PPD and extended door-opening (–0.2901), with significantly lower PPD during hours of extended door-opening (55.96% versus 68.88%, p < 0.001). CO₂ levels in this ward were also significantly lower during hours of extended door-opening (491 ppm versus 508 ppm, p = 0.0028).

The surgical ward showed minimal correlations between extended door-opening and both PPD (–0.0357) and CO₂ (0.0546), with higher CO₂ levels during hours of extended door-opening (861 ppm versus 814 ppm), although this difference was not statistically significant (p = 0.0638).

3.7 THERMAL IMAGERY

Figure 11 presents thermal images of exterior and interior hospital spaces captured across three study hospitals on 16 August 2024. Images were taken at different times of day (09:57, 08:48 and 12:40, respectively) and are therefore indicative rather than directly comparative.

Figure 11

Image A shows the exterior entrance of Kivunge Hospital, where mean surface temperatures of concrete-paved areas were 7.4°C higher than adjacent landscaped areas under similar solar exposure. Image B shows the ground-floor waiting area at Mwera Pongwe Hospital, which indicates minimal direct solar gains to interior surfaces. Image C shows the ground-floor ward at Chumbuni Hospital at 12:40. Elevated surface temperatures are visible at the aluminium window frame (Px2) and on the surface of the adjacent unoccupied ward bed (Px3), coinciding with direct solar radiation entering through the unshaded window.

3.8 SIMULATIONS OF WINDOW SHADING ADAPTATIONS

Figure 12 shows simulated direct sunlight exposure on windows at Kivunge Hospital for the north- and east-facing façades under existing conditions and with proposed shading adaptations. On the north façade, adaptations include combined horizontal and vertical side fins; on the east façade, vertical fins are applied to first-storey windows, while horizontal fins and strategic tree planting are proposed in front of upper windows. All shading fins project 300 mm from the façade plane. Together, these adaptations reduced simulated direct solar radiation on windows by 66% to 73%.

Figure 12

3.9 INPATIENT PERSPECTIVES

The majority of patients and relatives (n = 7/11, respectively) reported the presence of mosquitoes inside the hospital wards. Feedback on thermal comfort inside the wards varied with ward design, with patients in larger wards with more windows reporting better comfort and some voicing strong complaints around overheating. Patients and relatives described closing windows and doors as the only protective measure from mosquitoes in the wards (n = 6/11, respectively), with some also mentioning covering the body with sheets, use of personal repellent and ceiling fans preventing mosquito landing. Ceiling fans were also frequently described as effective in cooling the wards, though their efficacy was limited by their placement in the centre of the room rather than above the beds, their limited number, noise and variation in patient preference. Personal bed fans were suggested as a strategy to improve thermal comfort and the opening of windows. No patients specifically linked thermal discomfort with increased window and door-opening. Selected interviewee quotations are provided in Table 5.

3.10 REASONS FOR WINDOW- AND DOOR-OPENING IN WARDS

Participants in a brainstorming workshop identified 13 distinct reasons for windows and doors being left open in hospital wards (Table 6). The most frequently cited reason was air circulation, fresh air or ventilation (n = 22), followed by access and movement of patients, staff, equipment or for emergencies (n = 11). Reducing heat and reducing bad smells or odours were cited equally (n = 8 each). Broken or missing screens, windows, doors, locks or door closers were also commonly raised (n = 7), alongside low awareness or education around window screen use and closing practices (n = 5) and habitual or behavioural factors such as forgetting to close windows and doors (n = 5). Less frequently mentioned reasons included cleaning activities (n = 4), increasing light (n = 2) and disease transmission or patient safety concerns (n = 2).

4.1 THERMAL COMFORT, CO₂ LEVELS AND DOOR-OPENING

All spaces in the four study hospitals, except the air-conditioned office in Jitimai Hospital, failed to meet ASHRAE 55 thermal comfort recommendations for the majority of the study period. Thermal comfort was worst in the ward spaces compared to the waiting, clinic and office spaces. Inpatient interview feedback on thermal comfort in the wards varied from patients considering them as comfortable to others raising strong complaints about overheating (Table 5). This study was conducted during the cooler months of the year (May and June) and thermal discomfort is likely to be higher in the hottest months (December to February) (Adam et al. 2020).

Multiple hours were observed where uncomfortable PPD values overlapped with high CO₂ levels in the wards and some office and clinic spaces. Many of these hours also coincided with commonly reported peak mosquito biting hours found in the literature for Aedes, Anopheles and Culex species and illustrated in the heatmaps shown in Figures 9 and 10. These hours were typically at night, potentially increasing the risk of patients opening windows or doors to improve thermal comfort, thereby releasing a plume of high-concentration CO₂ and odour molecules detectable by host-seeking mosquitoes (Jatta et al. 2021). Night hours also coincided with increased occurrence of door-opening. However, the statistical analysis did not provide a clear correlation between door-opening and PPD or CO₂ levels, suggesting that door-opening may be due to factors described in interviews (Table 5) and the workshop activity (Table 6), such as patients and staff frequently entering or leaving the rooms or cleaning. Factors related to thermal comfort were the most reported reason for windows and doors being left open.

In air-conditioned spaces, CO₂ levels likely build up as a result of split air conditioning units recirculating rather than replacing indoor air, and rooms being kept closed to maintain cooler temperatures (Shin et al. 2021). Occupants’ thermal sensation has also been shown to become more sensitive to heat as the CO₂ concentration increased (Shin et al. 2021). This factor is not captured in the PPD model used in this study.

The main entrance doors and windows to waiting spaces tended to remain open during the day. While this reduces the façade’s effectiveness as a physical barrier against mosquitoes, it helps maintain CO₂ levels consistently lower in these spaces, potentially reducing the likelihood of mosquitoes finding hosts using CO₂ cues alone. These spaces are also most used in the daytime, when expected peak mosquito biting hours for Culex and Anopheles mosquitoes are fewer. This underscores the complex balance in designing passive buildings for tropical climates with MBD risks, between maintaining air permeability for thermal comfort and providing a physical barrier against mosquitoes. The improved PPD levels in waiting spaces compared with wards are likely influenced by their deep plan, increased ventilation, lack of direct solar radiation and use of high thermal mass materials.

4.2 HOSPITAL BUILDING DESIGN CONTRIBUTING TO THERMAL DISCOMFORT

Higher PPD and CO₂ levels in the wards are likely due to lack of ventilation and solar heat gains. The wards’ spatial arrangement, with openings typically on only one side, prevents cross-ventilation. This was worsened on the ground floor of Kivunge Surgical Ward (see Figure 10A), where the only operable windows open onto the internal ramp. Several other spatial factors in the wards impacted patient thermal comfort. Thermal images and photos shown in Figure 11 and interview feedback demonstrate overlapping design factors contributing to thermal discomfort:

• ward ceiling fans located in the room centre, away from beds, reducing their efficacy;

• lack of privacy screens for patients to remove clothes to improve comfort;

• ward window blinds with reduced effectiveness when air flows through the window with the potential to create a heat trap inside the room (Hu et al. 2023);

• some beds receiving direct solar radiation during parts of the day;

• heat transfer through aluminium ward frames without thermal breaks;

• large surface area around building using heat absorbing hard landscaping like concrete paving with minimal shading from trees;

• exposure of building façade to solar radiation.

At the time of the study, bed nets were not used in the wards. While bed nets could reduce MBD transmission risk, they also reduce airflow by about 60% in an enclosed space (von Seidlein 2012) potentially further reducing patient comfort (Bradshaw 2010).

4.3 SHADING ADAPTATIONS TO IMPROVE THERMAL COMFORT

Solar radiation simulation results in Figure 12 show that windows, particularly on the ground floor of the north façade and east and west façades, experience high levels of direct sunlight. A systematic review by Hu et al. (2023) identifies preventing solar gains through windows as the most effective method to reduce temperature and cooling demand in hot climates. The review also highlighted the importance of cross-ventilation in hot, humid climates like Zanzibar. Figure 12 also demonstrates how simple window shading strategies, such as overhangs, horizontal shading fins and strategic tree planting, can significantly reduce direct solar radiation through windows. Selective planting around the building would also shade ground surfaces receiving high solar radiation with the potential to reduce surrounding air temperature (Gillner et al. 2015). Interviewees also suggested the installation of more fans or personal fans for each patient. These adaptations would need to be tailored to each hospital based on surrounding shading and orientation to optimise their shading effect.

4.4 STUDY LIMITATIONS

The results should be interpreted cautiously as highlighting thermal discomfort as a factor may increase the likelihood of window and door-opening, potentially providing entry points for mosquitoes, and warranting further investigation.

Fanger’s PPD model was chosen over the Adaptive Thermal Comfort model, as it was considered appropriate for the hospital spaces where users have little control over their environment or may be incapacitated due to illness. PPD results are based on several assumptions such as internal air speed, clothing level and metabolic rate, which will vary and differ from real-world conditions across the areas of the hospital included in this study. PPD may overestimate thermal discomfort in this context and adaptive comfort models, which are better suited to naturally ventilated buildings, would likely indicate lower levels of discomfort. It should also be noted that environmental and interview data were collected at different times of year, meaning that indoor temperatures may reflect varying seasonal weather conditions.

Window-opening data were not available for this study. Sensors were installed on ward windows but failed to record reliably owing to the multipanel sliding window configuration. Door-opening was only monitored in one hospital. These factors limited the assessment of window- and door-opening as creating entry points for mosquitoes.

An indicative threshold above which mosquitoes can detect CO₂ was estimated at 530 ppm, based on the lowest CO₂ levels detected by mosquitoes as reported in the literature (108–133 ppm; Grant & O’Connell 2007) plus average atmospheric CO₂ levels (419–421 ppm; CenCO2PIP 2023). This threshold is still theoretical and has not been validated through studies. As the average CO₂ concentrations captured by a single sensor in this study do not capture the localised plumes near openings thought to primarily drive mosquito host-seeking behaviour (Jatta et al. 2021), and CO₂ concentrations vary temporally and spatially within rooms, this threshold should be interpreted cautiously rather than a definitive risk threshold.

The statistical analysis was exploratory in scope, and effect sizes and confidence intervals were not computed. Future studies with larger datasets should employ more comprehensive statistical modelling to better quantify the magnitude and uncertainty of associations between door-opening behaviour and environmental conditions.

4.5 CONTRIBUTIONS AND SUBSEQUENT RESEARCH

The 48-day continuous monitoring across 25 spaces in four hospitals contributes to the limited body of empirical indoor climatic data from health care settings in sub-Saharan Africa (Beauchamp-Davies et al. 2026; Roberts et al. 2022). To the authors’ knowledge, this is also the first study to empirically link thermal discomfort in health care settings with increased risk of mosquito entry.

Subsequent entomological surveillance using BG-Pro traps and Prokopack aspirators, conducted as part of the ongoing follow-up study, found mosquitoes present in wards despite screened windows and doors, supporting observations and interview data regarding the presence of mosquitoes inside the hospital despite the presence of window screens.

Together, these findings informed a set of co-created interventions targeting both thermal comfort as a risk factor for mosquito entry and therefore MBD transmission. These interventions included: personalised bed fans for each patient; extractor fans above ward doors to draw in cooler corridor air; self-closing door mechanisms and door brushes to seal floor-level gaps; repair and modification of window screens to prevent unscreened panel-opening; and façade shading using trees and climbing plants (Sloan Wood et al. 2025).

While the four study hospitals and focused analysis of Kivunge are not representative of all health care infrastructure in Zanzibar or the wider region, they share the same design as the nine district hospitals constructed in this period, suggesting that these findings and interventions are likely transferable to the remaining five. Variation in MBD transmission risk and the appropriateness of specific adaptations may nonetheless occur depending on site context, surrounding environment and the preferences of individual hospital communities.

4.6 FUTURE RESEARCH

A longer-term follow-up study is currently underway to address some limitations of this exploratory study as part of the MBD-Free Project. This will include:

• environmental data across a two-year period to account for seasonal changes;

• weather data from on-site weather stations to adjust for external variations;

• mosquito collection from different hospital spaces to assess species composition and host-seeking behaviour;

• semi-structured interviews and observations to gather qualitative data on hospital users’ experiences, comfort levels and behaviour in relation to the occupied spaces post intervention.