2.1 Eligibility criteria

Our inclusion criteria were based on the Population, Exposure, Comparator, and Outcomes framework [18]. Population: Studies were eligible when they focused on adult, professional or amateur athletes, performing in a real competitive setting. Exposures: Studies had to report objective air pollution measures the day or previous days before the competition(s). Comparators: Studies were eligible if they provided information on competition level, sex, type of performance, varying concentrations of air pollution exposure. Outcomes: Studies were eligible irrespective of objective measure of sport performance collected during a competition (e.g., race time, number of scored goals). The designs of included studies were observational studies quantifying the exposure-outcome relationship. Studies were excluded if they meet the following criteria: laboratory-based studies, indoor sport competition, performance(s) done outside a competitive context, subjective measure of air pollution, or examining volatile organic compounds, polycyclic aromatic hydrocarbons, or airborne metals. Some indirect performance outcomes were not included such as foul, yellow cards, corners, because they are more associated with referees’ decisions or tactical choices (e.g., percentage of possession, red card, long passes, corner during a football game) [19,20].

2.2 Search strategy

We conducted online research in the following databases: Pubmed, CINAHL, SportDiscus, GreenFILE, GeoRef, Scopus, JSTOR, for relevant publications up to April 2024, and then updated until December 2025. To reduce the risk of publication bias, three databases dedicated to gray literature were used (Open Grey, SportRxiv and ProQuest Dissertations). Furthermore, to find unpublished studies, we used two additional strategies. First, we e-mailed the first authors of the included published papers to know whether they possessed any unpublished data or were aware of unpublished studies by other researchers. Second, studies were sourced from six relevant reviews [5,21–25]. The search strategy was adapted for each database (further details are available in S1 File). Titles and abstracts of the initially identified articles were scanned by three reviewers for exclusion criteria (GC, QM, PB). Then, full-text articles were screened by GC, PC and PB.

2.3 Data collection process

Data were extracted on a pre-piloted spreadsheet by one author (PB) and cross-checked by six authors (QM, GC, QM, VB, CP, AO). All relevant studies were scrutinized to extract the following data: Authors, year of publication, study location, sample size, age, sex ratio, study design, air pollution outcome(s), duration of exposition, method of air pollution exposure assessment, nature of sport performance outcome(s), category of sport performance (endurance, speed/power, technical), number of competition examined, category of sport, elite level (based on definitions presented in [26]), weather confounders measured and adjusted for, other relevant confounders, multi-pollutant models, nature of previous short term exposition (e.g., lag in terms of hours, days), statistical analysis conducted, use of non-linear approach, and author’s conclusion.

2.4 Outcomes

The respective associations between particulate (PM_2.5, PM₁₀, PM_Coarse), gaseous (O₃, SO₂, NO₂, CO, black carbon, and index (Air Quality Index [AQI]) of air pollutants and sport performances were examined. Sport performances are described in supplemental materials (see S2 File), they include, for example, race time, pass accuracy, or first serve successful rate in tennis. Performance categories were assigned according to the physiological and technical demands of each sport (e.g., marathon = Endurance). For football and soccer, categories were assigned based on specific match-performance metrics; for example, number of passes was classified as Technical, distance covered in high-speed zones (11–13.99 km/h) as Speed/Power, and total distance covered as Endurance.

2.5 Risks of bias

A validated tool to assess the risk of bias for studies examining the association between air pollution and health/performance outcomes is missing [27]. We therefore used the six following criteria from a previous investigation [28]: “Is the sample generalizable to the population of interest?”, “Did the study a priori quantify sample and power?”, “Was missing data addressed and tested?”, “Was a dose-response relationship seen in any outcome?”, “Did the study design or analysis account for important confounding and modifying variables?”, “Did the study design or analysis adjust or control for other environmental exposures that were anticipated to bias results?”, “Did the study conclusions match the results?”. For each criterion, we indicated whether study met the requirement (i.e., binary outcome, yes/no), or not applicable. The risk of bias assessment was conducted by PB and GC, then checked by OA and AJR. Any disagreements were resolved by discussion.

2.6 Data synthesis

A narrative synthesis has been carried out to synthesize the findings from included studies. Harvest plot matrices presenting the associations between each air pollutant and performances categories (endurance, speed/power, technical) to which association was supported (negative, positive, not significant) have been generated [29]. Additionally, a methodological quality score (i.e., the inverted weighted mean of risk of bias) was assigned to each reported association in the harvest plots. These plots were used to facilitate the interpretation of findings and to illustrate the heterogeneity of the reported results.

2.7 Meta-analysis

Pearson’s r or (un)standardized beta were used as effect size, and transformed in Cohen’s d. All effect sizes were pooled for a fixed increment in pollutant concentration (per 10 μg/m³), other reported quantities or units (such as parts per billion and parts per million) were converted when appropriate (see details in S1 File). We assumed a linear association between air pollutants and performances.

When results were available from both single- and multi-pollutant models, only the multipollutant model results were included because mixed exposure is more representative of real-life situations. Magnitude of effect sizes (d) were interpreted as small >0.20, medium >0.50, and large >0.80 [30]. Meta-analyses employed a robust variance estimation to account for dependencies of having multiple effect sizes come from each study. Statistical dependency is very common in environmental research [28]. This method allows simultaneous analysis of multiple effect sizes per study with accurate estimates and standard errors, even when information on the covariance of these effect sizes is unavailable. As recommended [31], a meta-analysis was performed only if 4 or more associations were reported from different studies. Separate analyses were carried out for each pollutant. When the number of available estimates was equal or higher to 4, a subgroup analysis for each performance category (i.e., endurance, speed/power, technical) was also performed.

Heterogeneity was quantified with the I² statistic [32]. We also report the method of moments estimate of τ², the between-studies variance component, as a measure of between-study heterogeneity in effect sizes. If an association below p < 0.10 was found a prediction interval was provided, possible outlier and publication bias were also explored. The prediction interval provides a range within which we can expect the associations of future studies to fall based on the current evidence.

Traditional influence or publication bias analyses are not implemented in the robumeta package used to perform the robust variance estimation. To overcome this limitation, we followed the method used in previous meta-analyses [33]. First, univariate outliers were inspected with the Grubbs test on all effect sizes. Then, we conducted publication bias analysis using the PublicationBias package and calculated corrected effect sizes assuming significant results were 2 times more likely to be published than non-significant results. These effect sizes conservatively reflect estimates of publication bias greater than those found in a meta-meta-analysis.

2.8 Sensitivity analyses

As recommended [34], a set of sensitivity analyses were also carried out to verify the robustness of ours estimated effect sizes. Results from the robust variance estimation meta-analyses were compared with those from multi-level meta-analyses [35].

All analyses were carried out in R 4.3 using the robumeta package [36] for the robust variance estimation meta-analysis, the metafor package [37] for the multi-level meta-analysis. Analyses were performed by PB and independently replicated by GC and LH. The dataset and annotated R code are available as supplemental materials on the Open Science Framework page of the project (see: https://osf.io/cb6fu/).

3.1 Study selection

The initial electronic searches identified 1172 references, of which 107 were duplicates. After a review of titles and abstracts, 1019 were excluded because they did not meet all inclusion criteria. Hence, 46 full-text articles were assessed for eligibility, and 21 were further included in the review (see Fig 1). Additionally, two studies were provided by authors who were contacted. No other additional studies were identified through hand search of relevant articles. The corresponding authors of articles with missing information were contacted.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart.

https://doi.org/10.1371/journal.pclm.0000988.g001

3.2 Study description

We conducted three complementary knowledge synthesis approaches. First, we extracted key methodological characteristics and findings from each included study, which are presented in Table 1 and S2 File. Second, we graphically represented the associations between air pollution measures and the three categories of performance (endurance, technical, and speed/power) using a Harvest Plot. At this stage, the number of represented associations exceeds the number of included studies, because several articles reported associations for multiple performance outcomes (e.g., distance covered and pass accuracy in soccer) or for distinct participant subgroups (e.g., elite vs. amateur, men vs. women). The study by Xu et al. was not included in Fig 2, as the reported outcome (number of goals scored) could not be unambiguously classified within our predefined performance categories. Third, we performed a series of meta-analyses restricted to studies providing the statistical information required to compute standardized effect sizes and their variances. Studies that did not meet these criteria are highlighted by grey rows in Table 1 and were not pooled quantitatively.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Short summary of included studies.

https://doi.org/10.1371/journal.pclm.0000988.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. World map of included studies.

Notes. Multi-center studies are represented here at the national (colored areas) and local (city circles) scale. The map was plotted using the R package rworldmap (Natural Earth data basemap).

https://doi.org/10.1371/journal.pclm.0000988.g002

Table 1 provides a summary of the 23 included studies. More details are presented in S2 File. The oldest article retrieved was published in 1967 and the most recent one 60 years later, in 2025. Football was the most studied sport (n = 10) follows by short (e.g., 5 km running) and long endurance sports (e.g., marathon, ironman, n = 10). Performances data from US football, tennis, and track and field practices (i.e., sports including running, jumping, and throwing practices) were examined in one study, respectively. All performances were characterized at the individual level, except for two studies examining collective performances [38,39]. Sixteen studies were conducted with data from elite athletes (e.g., top 3 marathon finisher, professional football player), one only with amateur athletes, and six with both. The number of performances reported within-study ranged from 20 to 2,564,811.

Fig 2 illustrates the countries and cities in which the studies were carried out, China was the most represented country. Most studies were conducted in high-income countries according to World Bank classification (62%), the remaining studies were conducted in low- and middle-income countries (38%).

Regarding the measures of air pollution, data were provided by the environmental authorities of the cities or countries concerned. Analyses from 17 studies (73%) were based on local air pollution stations. Meteorological conditions were not always detailed, 18 studies included weather related variables in their models (most frequent were: temperature, humidity and wind speed). Eleven studies (47%) performed multipollutant statistic model. Additionally, some data of pollutants are also missing in multicenter studies. For instance, some pollution sensors have been moved or did not measure all the same parameters, as is the case for PM₁₀ in the study led by Helou et al. (2012) [48], which is not available for all cities. In general, the risk of bias was relatively low, the items assessing the missing data management and a priori power analyses had the higher rate of missing information. Table 2 presents the scoring of studies’ risk of bias.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Methodological quality of included studies.

https://doi.org/10.1371/journal.pclm.0000988.t002

3.3 Type of air pollutants studied

The pollutants reported in the studies were PM_coarse, PM₁₀ and PM_2.5, O₃, NO₂, CO and SO₂. AQI was also considered among included studies. Five studies classified air quality during events into categories ranging from very good or good to hazardous or very poor, based on the pollutant with the worst AQI value [20,40–43].

3.4 Concentrations of air pollutants and sampling frequency

The details of PM and O₃, CO, NO₂ m AQI and SO₂ concentrations are given in supplemental materials (S1 File). The hourly values of local pollution data were used by all authors, except for Fleury et al. [44]. The data were mostly reported as monitored and averaged over either the event time for each athlete, the time a team spent on the field or the duration of the match [20,41,42,45–48]. El Helou et al. (2012) used an average of 4 hours from the start of each marathon [49], and Humphreys et al., (2023) calculated average air pollutants from 12:00–18:00 during NFL games [47]. Hodgson et al., (2022) considered two fixed air quality values between 10:00 and 12:00 and between 10:00 and 15:00 depending on the group of runners (elite vs amateur) [50]. Some authors used daily averages during the day of the event [16,43,51], and Marr et al., (2010) used 2-hour averages for gaseous pollutants during the race and 24-hour averages on the day of the race for PM [10]. Zacharko et al. (2023) used three measurements of air pollution during the game, i.e., at the beginning, during the break and at the end of the game [52]. Finally, few authors characterized exposure to air pollutants prior to the sporting event until the day of the meet/match. For instance, Cusick et al (2023) averaged exposure to air pollutants 20 days before the event [40], Mullins (2018) 7 days before the start [53], Wayne et al., (1967) 3 hours before the event [38] and Qin et al., (2022) 3 months before the event [54].

3.5 Weather conditions

The range of temperature reported varies from -12.2°C to 36.1°C, relative humidity from 5.6 to 100% and wind speed from 0 to 68 km.h^-1 (see S1 File. Table 5). Most data sources were obtained from official meteorological services while a few authors did not specify the source of the weather data [53,54] or use specific data sources such as the one from the National Football Ligue [47]. Hourly weather conditions were mainly averaged over the duration of the sporting event [10,40,41,45,47,50,54,55]. As for air pollutants, El Helou et al. (2022) used the 4 hours average after the start of each race [49], and Hodgson et al. (2022) averaged the values from 10:00–12:00 and from 10:00–15:00, depending on the group (elite or amateur). Picchio et al. (2024) [16] used the temperature at 15:00 two meters above the tennis court and Zhang et al. (2024) [43] used a daily average.

3.6 Sport related covariates

Eleven studies included sport specific related variables in their multivariate models. For instance, Cusik et al. (2023) [40] added the 5 km running athlete’s personal record, and days since the last race. In football, player position, stadium attendance, home advantage, team market value, play time, and opponent strength have also been used as covariates.

3.7 Amateur versus elite status

Studies examining how the relationship between air pollution and performance varies by athlete status (amateur vs. elite) have found mixed results and the literature is rather inconclusive. For example, high AQI levels were linked to longer running times for both amateur and elite runners [41]. High O₃ levels were associated with slower running times, but only for the slowest runners and elite female runners (41]. In contrast, elevated O₃ levels negatively impacted running times for elite runners (Top 20) but not running time for the slowest runners [56]. High NO₂ levels correlated with longer race times for all runners, including the top quartile of finishers [49]; however, this effect was not observed in analyses focusing on the top 1% of runners. Another study analysed the effects of O₃ and PM₂.₅ on both professional and amateur Ironman athletes. Authors found no association between overall race or running times and O₃ and PM₂.₅ levels across all Ironman athletes. However, a significant negative association was identified between O₃ levels and bicycle riding time, but only among professional athletes [51]. Regarding PM₂.₅, a percentile analysis of finish time among 140 US marathons between 2003 and 2019 found greater effects of PM₂.₅ among median-and-faster runners [44].

3.8 Sex differences

The evidence regarding sex-specific differences in the effects of air pollution is inconclusive. In terms of sex differences, five studies found comparable negative associations between air pollution levels and performance for both male and female athletes. Specifically, AQI [41], PM₂.₅ [44] and NO₂ [48] were associated to longer marathon race times for both male and female athletes, and O₃ levels were associated with decreased athletic performance for both the slowest amateur runners and elite athletes [49,52]. Higher PM₁₀ concentrations were associated with longer race times in female, but not in male, athletes in one study.

3.9 Associations between outdoor air pollution and sport performances

3.9.1 Harvest plots.

In general, a consistent pattern of association was found for AQI, PM_2.5, and NO₂ with endurance performances (see Fig 3). Only six studies examined the relationship between air pollution and speed/power performance outcomes, making it difficult to identify a clear pattern of association for these outcomes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Harvest plots of tested associations between air pollution and sport performances.

Each bar represents a tested association; the height of the bar represents the article’s methodological quality, and the colour of the bar represents the number of used observations. Findings from Xu et al (2024) were not represented in this plot because they examined the number of scored goals. NA = not applicable.

https://doi.org/10.1371/journal.pclm.0000988.g003

Air quality index: Of the fourteen-association tested between AQI and endurance-related performances, height showed a significant negative association, with a low risk of bias. No significant association was found between AQI and technical performances, with six of nine associations showing no significant association (moderate risk of bias).

PM₁₀ and PM_2.5: Studies examining the association between PM_2.5 and endurance performances showed a relatively similar pattern, with twenty height associations reporting no significant effects (ranging from low to high risk of bias). The evidence from ten tested associations between PM_2.5 and technical performances was inconclusive. Similarly, findings on the association between PM₁₀ and performance outcomes did not indicate a consistent pattern.

O₃: No clear pattern of association was found between O₃ and endurance performances. Eleven associations were significantly negative (low risk of methodological bias), while nineteen associations showed no significant effects (ranging from low to high risk of bias).

NO₂: No significant association was observed between NO₂ and endurance performances, with twelve reported no significant associations moderate to high risk of bias (see Fig 3).

SO₂, CO and PM_coarse: No significant association was found for SO₂ and endurance performances, with eight of nine associations showing no effect (moderate risk of bias). Due to the limited number of studies examining associations between CO, PM_coarse, and performance outcomes, it is challenging to establish a global pattern of results for these pollutants.

3.9.2 Meta-analyses.

All associations reported in the harvest plots are not included in the meta-analyses because of missing quantitative information. Three studies were excluded [42,45,52], and three others were only partially used [51,53,56]. In total, we used 150 out of the 225 associations retrieved (67%). Our estimates for AQI, CO, NO₂, O₃, PM₁₀, PM₂.₅, and SO₂ were based on the following proportions of reported findings from included studies: 26/28 for AQI (93%), 9/9 for CO (100%), 19/37 for NO₂ (52%), 32/56 for O₃ (58%), 14/32 for PM₁₀ (43%), 37/46 for PM₂.₅ (80%), and 11/13 for SO₂ (85%). The meta-analyses incorporated between 280,000 and 2.7 million individual observations.

Regarding pooled and endurance performances, an analysis has been conducted for AQI, CO, NO₂, O₃, PM₁₀, PM₂.₅, and SO₂. For technical performances, a subgroup analysis has been carried out for PM_2.5. No statistically significant associations were found using robust variance estimations. Multi-level meta-analysis models were carried out as sensitivity analyses. Two statistically significant and negative effect sizes were found for O₃ and all performances (d = -0.33, 95%CI: -0.65 –0.02) and endurance performances (d = -0.54, 95%CI: -1.02 -0.04). No statistically significant associations were obtained for the remaining five air pollutants and performance outcomes. Results from the robust variance estimation are available in Table 3. Further details about the sensitivity analyses as well as forest plots are available in supplemental materials (see S3 File).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Findings from RVE meta-analyses.

https://doi.org/10.1371/journal.pclm.0000988.t003

4.1 Principal findings

This systematic review and meta-analysis offer the most comprehensive evaluation to date of observational studies examining the associations between short-term exposure to ambient air pollution and sports performance. Overall, we found no clear evidence of a negative association between outdoor concentrations of AQI, PM_2.5, and NO₂ and sports performance in our harvest plots and meta-analyses. Short-term O₃ exposure however was associated with lower performances of all types (moderate effect size) and endurance performances (small effect size) within the multi-level meta-analyses. Meta-analyses for CO, PM₁₀, and SO₂ showed non-statistically significant results and the associations presented in harvest plots were inconclusive.

Our findings for O₃, based on the multi-level meta-analysis, align with a previous systematic review by Hung et al. (2022) [7], which concluded that O₃ exposure during physical exercise in laboratory settings can impair endurance performance [7]. A key distinction between our findings and this previous systematic review lies in the concentration levels tested: laboratory settings typically used O₃ concentrations ranging from 118 to 1472 µg/m³ (60–750 ppb), whereas real-life measurements during sports activities rarely exceeded 111 µg/m³ [39].

Regarding PM, PM_2.5, and PM₁₀, our results do not align with the conclusions of previous reviews which suggested that short-term exposure to PM impairs performances in laboratory setting [5,6]. Only a limited number of studies reported associations between PM₁₀, PM_coarse and performance outcomes in our review. In our review, 35 of the 45 reported associations between PM_2.5 and performances, originating from 13 studies, were non-significant. Across the 13 studies examining PM_2.5, 25 of the 36 reported associations were non-significant. Because multiple associations from the same study share sources of methodological and contextual bias (sample, exposure assessment, design) and cannot be treated as statistically independent, this descriptive count is provided for transparency only and should be interpreted alongside the RVE meta-analytic estimate reported in Section 3.9, which formally accounts for within-study clustering. Taken together, these results indicate that short-term exposure to particulate matter does not appear to reduce performance in real-life competitive settings, in contrast to the effects observed under controlled laboratory conditions [5,6]. This discrepancy may be explained by differences between the ranges of pollutant concentrations tested under laboratory conditions and those measured in real-life settings. Indeed, Hung et al (2022) [7] reported laboratory exposure concentrations ranging from 13.8 to 250 µg/m³ for PM₁₀ and from 9.7 to 302 µg/m³ for PM_2.5. In contrast, the concentrations observed in the studies included in the present review were substantially lower, ranging from 12.4 to 104 µg/m³ for PM₁₀ and from 7.2 to 46.8 µg/m³ for PM_2.5.

These findings should be interpreted with caution, especially in cases of severe pollution. For example, during the 2014 Beijing Marathon, visibly high PM levels (e.g., smog) may have influenced athlete behavior, leading some to withdraw, wear masks, or forgo participation altogether [57]. This may mean that only less sensitive athletes and/or already locally exposed residents participated in polluted conditions. Similarly, respiratory symptoms have impaired performance during events affected by mega-fires, such as the 2019 Australian Open tennis tournament, held during Australia’s “Black summer” [58]. In some cases, events may be postponed due to hazardous air quality, as during the 2023 Canadian mega-fires [59]. While it is safer to cancel sporting events when pollution levels are too high, this impedes the investigation of high levels of air pollution on performances in real-life competitions.

Our analyses indicate that no definitive conclusions can be drawn regarding the associations between CO, NO₂, and SO₂, due to the limited number of tested associations and the resulting statistical imprecision. This is consistent with laboratory studies, where the evidence for a negative effect of these pollutants on endurance performance was judged to be low [7,8]. As for AQI, our findings cannot be directly compared with laboratory investigations, since AQI is a composite index that cannot be replicated under controlled conditions. Additionally, a potential non-linear relationship between AQI and performance [41] may partly explain the lack of statistically significant associations observed in our meta-analyses.

4.2 Strengths and limitations

This review has several strengths. It used a reproducible, systematic and standardized approach to produce both qualitative and quantitative syntheses of the findings, including data from 47 countries across all continents. Additionally, our literature synthesis is the first to include AQI and five other air pollutants. Our meta-analyses were conducted with a large number of observations, which increases the robustness of our findings. Finally, our work is the first to summarize the associations between air pollution and sports performance in real-life contexts.

Regarding limitations, because we focused on observational studies, causal inference about the relationships between air pollution and sports performance is limited. Second, most investigations used air pollutant monitoring data or modeled concentrations instead of hyper-local or personal exposure measurements, which may misrepresent actual exposure and potentially underestimate negative effects [60]. Third, half of all studies reported adjusted estimates, the number of covariates varied among different studies and important covariates were not considered (e.g., wind) in all studies or characterized with different approaches across studies. Fourth, early warning systems for air pollution are increasingly implemented in urban areas, potentially leading to canceled sports events or the adoption of coping strategies by athletes (e.g., medications, antioxidant supplementation). However, none of the reviewed studies reported on these systems or strategies, so their potential impact could not be accounted for. Five, the number of associations available between air pollutants and speed/power performances was insufficient to perform meta-analyses. Sixth, the associations presented in Fig 2 may be influenced by the number of sub-analyses conducted within the included studies. For example, the same association could have been examined separately for the overall sample, by sex, or among top-three finishers.

4.3 Future research

The reviewed literature reveals considerable methodological and outcome variability, underscoring the importance of more standardized approaches in future research. First, researchers could develop large-scale, collaborative projects to examine the short-term associations between air pollution and a specific set of comparable performance indicators within the same sport (e.g., home runs or slugging percentage in baseball). Further, minimal research has been conducted on technical performance, and no studies have examined motorized sports, cycling, bat-and-ball games, or archery/shooting sports. Additionally, more studies focusing exclusively on female [61] and amateur athletes are needed to explore their specific vulnerability to air pollution. Second, we need more direct measurement of air pollution and weather outcomes implemented directly in sports settings (e.g., stadiums). Dynamic air pollution monitoring systems could be used during sport competition [60]. It combines several stationary air pollution monitors positioned along a marathon course with a mobile monitor devices mounted on a bike. This approach would enable researchers to obtain highly granular estimates of pollution exposure among both elite and recreational runners, while also identifying localized sources and temporal spikes in air pollution levels. The current development of compact wearable air pollution sensors (designed to be worn on the wrist) provides promising opportunities for future citizen science initiatives conducted during endurance events [62]. Third, future studies should examine the effects of air pollution exposure not only on the day of the competition (acute exposure), but also a few days before for visiting athletes, or longer periods for home athletes. For national to international events, it would also be helpful to modelize the chronic exposure of athletes during their daily lives. Two previous studies have suggested that soccer teams unaccustomed to high levels of air pollution had lower performance when competing against teams regularly exposed to higher pollutant concentrations [19,54]. Fourth, potential non-linear associations between air pollution and performance should be systematically examined. Specifically, both official and empirically derived pollutant thresholds should be considered. For instance, the official PM10 alert threshold in Germany is 70 µg/m³ [63], whereas a data-driven threshold linked to reduced performance in elite German soccer players was 55 µg/m³ [48]. Integrating both approaches can help sports stakeholders and coaches establish sport-specific pollutant thresholds for amateur and elite athletes. Fifth, more sport-specific, contextually grounded mechanistic studies are needed to better understand the potential pathways through which air pollutants affect performance outcomes. For example, high air pollution levels have been linked to lower spectator attendance and increased frequency of referee errors [64], which could partly explain the observed effects of air pollution on sports performance.