Effectiveness of indoor air purification intervention in improving cardiovascular health: A systematic review and meta-analysis of randomized controlled trials.

Indoor air purifiers are increasingly marketed for their health benefits, but their cardiovascular effects remain unclear. We systematically reviewed and meta-analysed randomized controlled trials (RCTs) on the cardiovascular effects of indoor air purification interventions in humans of all ages. We searched Embase, Medline, PubMed, and Web of Science from inception to 22 August 2020. Fourteen cross-over RCTs (18 publications) were included. Systolic blood pressure (SBP) was significantly reduced after intervention (-2.28 (95% CI: -3.92, -0.64) mmHg). There were tendencies of reductions in diastolic blood pressure (-0.35 [-1.52, 0.83] mmHg), pulse pressure (PP) (-0.86 [-2.07, 0.34] mmHg), C-reactive protein (-0.23 [-0.63, 0.18] mg/L), and improvement in reactive hyperaemia index (RHI) (0.10 [-0.04, 0.24]) after indoor air purification, although the effects were not statistically significant. However, when restricting the analyses to RCTs using physical-type purifiers only, significant improvements in PP (-1.56 [-2.98, -0.15] mmHg) and RHI (0.13 [0.01, 0.25]) were observed. This study found potential evidence on the short-term cardiovascular benefits of using indoor air purifiers, especially for SBP, PP and RHI. However, under the Grading of Recommendations Assessment, Development and Evaluation framework, the overall certainty of evidence was very low, which discourage unsubstantiated claims on the cardiovascular benefits of air purifiers. We have also identified several key methodological limitations, including small sample size, short duration of intervention, and the lack of wash-out period. Further RCTs with larger sample size and longer follow-up duration are needed to clarify the cardiovascular benefits of air purification interventions.


Introduction
Air pollution is the leading environmental risk factor for ill health globally, estimated to account for 4.9 (Health Effects Institute., 2019) to 8.8 (Lelieveld et al., 2020) million deaths (largely from cardiovascular disease) per year. As people spend most of their time indoors, there have been widespread health concerns about indoor air pollution, particularly fine particulate matter (PM 2.5 ) (Klepeis et al., 2001). Apart from tobacco smoke, the primary sources of indoor PM 2.5 include domestic combustion of solid fuels (e.g., coal and wood) for cooking and heating in rural areas (Bruce et al., 2015) and the infiltration of ambient particulates in urban areas (Branco et al., 2014;Guo et al., 2013;Habre et al., 2013;Tong et al., 2016). In poor ventilation conditions, indoor PM 2.5 levels can build up to several times higher than outdoor levels even in the absence of solid fuel combustion (Ramachandran et al., 2003).
Ample evidence suggests that exposure to PM 2.5 is associated with excess risks of cardiovascular disease (CVD) and mortality (Newby et al., 2015;Rajagopalan et al., 2018). The postulated mechanisms include elevated blood pressure (BP), endothelial function impairment and systemic inflammation (Pope III et al., 2016). Correspondingly, there is a range of well-established biomarkers used in epidemiological assessments, including pulse pressure (PP), reactive hyperemia index (RHI), C-reactive protein (CRP), interleukin-6 (IL-6), and fibrinogen (Newby et al., 2015). It has been increasingly suggested that a modest reduction of ambient PM 2.5 exposure at a population level can result in substantial public health benefits, but this is based predominantly on observational studies instead of gold standard randomized controlled trials (RCTs) (Health Effects Institute., 2019; Wei et al., 2019). At the same time, the use of indoor air purifiers against PM 2.5 has received growing attention (Eggleston et al., 2005;Rajagopalan et al., 2018) and they are increasingly marketed as a health commodity, especially in populations where strong policy interventions against air pollution levels are not available, but the cardiovascular benefits of such interventions remain unclear.
In an earlier meta-analysis of intervention studies on the effects of using air purifiers on systolic (SBP) and diastolic (DBP) blood pressure, Walzer et al. reported a significant reduction in SBP .00, -0.89] mmHg) but a non-significant effect on DBP (-0.95 [-2.81, 0.91] mmHg) in the intervention group (Walzer et al., 2020). However, this meta-analysis included both RCTs and non-randomized studies, and an older and non specific risk-of-bias assessment framework was used. Two other recent qualitative reviews described a broader range of studies involving other CVD biomarkers, but they did not employ a systematic evidence quality assessment framework and no meta-analysis was conducted (Allen and Barn, 2020;Cheek et al., 2021). In order to more comprehensively and critically assess the cardiovascular effects of reducing PM 2.5 exposure through indoor air purification, we conducted a systematic review and meta-analysis following the well-established GRADE (Grading of Recommendations Assessment, Development and Evaluation) framework to synthesize and evaluate the RCT evidence on the effects of air purification intervention on SBP and DBP as primary outcomes and other cardiovascular biomarkers (e.g. PP, RHI, CRP) as secondary outcomes in humans of all ages. and Sally, 2011). Bibliographic references of all articles included after the screening of titles and abstracts were checked for additional studies. We also searched for unpublished trials registered in ClinicalTrials.gov using the terms "air filter" and "air purifier". This study has been registered at the Open Science Forum (https://doi.org/10.17605/OSF.IO/ F2R9M), and this report follows the Preferred Reporting Items of Systematic Reviews and Meta-analysis (PRISMA) guidelines (Liberati et al., 2009).

Eligibility criteria and study selection
The inclusion criteria were as follows: 1) study design: RCT; 2) intervention: any type of air cleaner/ purifier/ purification device that was used in indoor environments, including household, office, school, etc; 3) participants: humans, with no limitation on age or medical history; 4) outcomes: SBP and DBP as primary outcomes and any health outcomes related to cardiovascular health as secondary outcomes (for exploratory investigation); 5) full-length peer-reviewed studies; 6) language: English. All articles were first screened for title and abstract, then reviewed in full-text by two independent reviewers (XX and KHC) to evaluate their relevance, and any disagreement was forwarded to KFH.

Data extraction
XX and KHC independently extracted the data based on the double data entry requirement (JPT and Sally, 2011). Extracted data included citation, participant characteristics, study design, region, intervention details (e.g., type of air purifier, setting, washout period, duration, etc.), and information on air pollutants (PM 2.5 concentrations in control and intervention groups, and the reduction efficiency). Where possible, the means and standard deviations (SDs) of the reported health outcomes measured post-intervention and post control periods, and the mean differences (and corresponding SDs) between arms were extracted. If such information was not reported, SDs were calculated from standard errors, 95% CIs or ranges. If the study only reported geometric means, we converted the data to arithmetic means using an established method (Higgins et al., 2008). For studies that only reported percentage changes of CVD outcomes associated with the intervention, the mean differences between the intervention and control arms were estimated as the product of the baseline values and estimated percentage change. If the results were published in figures only, Web Plot Digitizer was used for data extraction (Rohatgi, 2020).

Methods for Meta-analyses
Meta-analyses were conducted for the outcomes that were comparably reported in four studies or more. Mean differences with 95% CI (post-intervention minus post-control values) were pooled using inverse-variance weighting, as such data were reported by most of the studies. Where different units of measure have been reported across studies, standardized mean difference was used. The heterogeneity across studies was assessed by I 2 statistic (Ioannidis et al., 2007). When I 2 < 25%, fixed-effect model was applied; otherwise, random effect model was used (Mantel and Haenszel, 1959). For outcomes reported in more than six studies, subgroup analyses were conducted to compare the pooled-estimates by baseline PM 2.5 levels (≤25 vs. >25 μg/m 3 ), intervention-PM 2.5 levels (≥10 vs. >10 μg/m 3 ), baseline blood pressure (SBP<120 vs. SBP≥120 mmHg), study setting (at home vs. at school), type Xia et al. Page 4 Sci Total Environ. Author manuscript; available in PMC 2021 October 01.
Europe PMC Funders Author Manuscripts of air purifier (physical-type vs. electrostatic or ionization), intervention duration (≤7 vs. >7 days), health condition of participants (healthy subjects only vs. mixed), level of risk of bias (with low risk or some concerns vs. high risk), and the location (China vs. others) of the study to help explore the heterogeneity (Borenstein et al., 2011). Leave-one-out analyses were performed to test the robustness of the pooled estimates. Funnel plots and Egger's regression were used to evaluate the risk of publication bias (Peters et al., 2006). All meta-analyses were conducted using the 'metafor' package in R version 3.5.3 (DerSimonian and Laird, 1986; R, 2018).

Risk of bias assessment
The risk-of-bias of the individual studies for each outcome was assessed using the Cochrane Risk-of-Bias Version 2 (RoB2) tool (Sterne et al., 2019). The RoB2 tool includes five domains relevant to the major sources of bias in RCTs, including risk of bias arising from (i) the randomization process (Domain 1), (ii) deviations from the intended interventions (Domain 2), (iii) missing outcome data (Domain 3), (iv) measurement of the outcomes (Domain 4), and (v) selective reporting (Domain 5). Each domain was assessed following standardized guidelines and determined to have "low risk of bias", "some concerns", or "high risk of bias". Each domain consists of a series of detailed signalling questions, which are well-suited criteria for a systematic assessment of risk of bias in RCTs (https:// www.riskofbias.info/welcome). For example, if the participants had the risk of being aware of their assigned intervention group during the study, Domain 2 would be determined as having "some concerns". Finally, based on a summary of the domain-level judgements, an overall risk-of-bias judgement with three final levels (i.e., low, some concerns, high) can be determined (see Appendix B in the Supplementary file). Any disagreement in the risk of bias assessment between XX and KHC was forwarded to KFH and resolved by discussion.

Certainty of evidence assessment
The certainty of the body of evidence for each health outcome was assessed using the GRADE framework (Higgins and Thomas, 2019). The assessment was based on outcome specific groups of studies instead of a judgement for each individual RCT. According to the GRADE guidelines, the certainty of evidence was categorised into four levels (i.e., high, moderate, low or very low). The initial certainty of a body of evidence for RCT starts at the "high" level (i.e., high confidence between true and estimated effect), and then it could be downgraded for five reasons -risk of bias, imprecision, inconsistency, indirectness, and publication bias (see Appendix C in the Supplementary file) -as they could cover most issues that bear on the certainty of evidence (Balshem et al., 2011).

Risk of bias
The risk of bias judgements and corresponding details for each domain of each included study are presented in Appendix D, and the outcome-specific RoB2 judgements for each domain of all included studies are summarized in Appendix E in the supplementary file.
The dropout rate ranged from 0% (in five RCTs) (Chen et al., 2015Chuang et al., 2017;Liu et al., 2018Liu et al., , 2020bMorishita et al., 2018;Shao et al., 2017) to 20% (overall mean=6.5%). As most of the studies involved relatively short-term interventions, participants were asked to stay in the air-filtered areas, keep the windows closed, and/ or avoid cooking or cleaning as long as possible. Participants in nine RCTs spent on average 85% of their time indoors (range: 74%-100%), indicating a high compliance (Allen et al., 2011;Chen et al., 2015Chen et al., , 2016Chen et al., , 2018Cui et al., 2018;Dong et al., 2019;Kajbafzadeh et al., 2015;Karottki et al., 2013;Li et al., 2017;Liu et al., 2018Liu et al., , 2020aPadró-Martínez et al., 2015;Shao et al., 2017). Eleven RCTs clearly described their double-blinding procedures, while three RCTs only reported the blinding of participants (Allen et al., 2011;Kajbafzadeh et al., 2015;Shao et al., 2017). In particular, twelve RCTs used a sham filter in the purifiers for the control periods, whereas the two RCTs with ionization air purifiers had the devices switched off (Dong et al., 2019;Liu et al., 2020a; two publication from one RCT) or the internal power supply wire severed (Liu et al., 2020a), which might have impaired the concealment leading to bias.
The leave-one-out sensitivity analysis demonstrated the robustness of the result for SBP (Appendix F in the Supplementary file).
Although the subgroup analyses found no statistically significant differences for BP reduction in relation to the pre-specified characteristics, there was a tendency of greater reduction in SBP in studies conducted among participants with higher baseline SBP (i.e.,
Although the subgroup analysis showed no significant differences for the selected characteristics (Appendix G Table S2 in the Supplementary file), the subgroups that observed somewhat stronger reductions in PP were the same as those for BP (e.g., RCTs conducted in household environments or those with longer intervention durations). Besides, the Egger's test for funnel plot asymmetry was non-significant (P = 0.50; Figure 4C).
Five RCTs measured RHI (Allen et al., 2011;Bräuner et al., 2008;Kajbafzadeh et al., 2015;Karottki et al., 2013;Weichenthal et al., 2013), an indicator for vascular endothelial dysfunction, and they reported inconsistent results. Because one RCT only reported the median and the 5 th and 95 th percentiles of RHI (Karottki et al., 2013), we conducted a meta analysis on the other four RCTs ( Figure 5B). Overall, there was a marginally non-significant improvement of 0.10 (-0.04 to 0.24) associated with air purification (I 2 = 26.6%). In the leave-one-out analysis (Appendix F in the Supplementary file), the removal of one RCT that used electrostatic air purifier (Weichenthal et al., 2013) led to a statistically significant improvement in RHI (0.13 [0.01, 0.25]). No statistical evidence for publication bias was found (P = 0.59; Figure 4D).

Effects on the autonomic nervous system-Only
four RCTs reported the effects of indoor air filtration interventions on the autonomic nervous system (Cui et al., 2018;Dong et al., 2019;Morishita et al., 2018;Shao et al., 2017), which precluded a meta-analysis. Two of them examined the changes in heart rate (HR): one observed a non-significant reduction of 1.47 (-3.72, 0.79) min -1 (Cui et al., 2018), and one found a significantly higher HR (mean ± SD: 92 ±12 min -1 in the true-filter group vs. 91±13 min -1 in the sham-filter group, P < 0.001) (Dong et al., 2019). Besides, three trials examined heart rate variability (HRV), but the indicators used were heterogeneous (e.g., high frequency [HF], low frequency [LF], LF/HF, the square root of the mean of the squared differences between adjacent normal-to-normal intervals [RMSSD], and the standard deviation of the normal-to-normal interval [SDNN]) (Dong et al., 2019;Morishita et al., 2018;Shao et al., 2017). One RCT that used ionization air purifiers reported significantly lower HRV indices during the true air purification period (P < 0.001) (Dong et al., 2019), whilst one observed non-significant negative effects (Shao et al., 2017) and another reported a non-significant improvement (Morishita et al., 2018).
Of the nine trials that measured CRP, three provided insufficient or incompatible data to be transformed to means and SDs (e.g., only reported the median [5 th percentile, 95 th percentile] or percentage change only) (Karottki et al., 2013;Li et al., 2017; Padró-Martínez  et al., 2015), so only the remaining six RCTs were meta-analysed. Overall, a non-significant pooled reduction was found (-0.23 [-0.63, 0.18] mg/L; I 2 = 48.1%; Figure 5C). The subgroup analysis suggested that RCTs with longer intervention duration (over seven days) tended to have a larger reduction in CRP (-2.78 [-4.53 Figure 4E).
The concentrations of MDA and 8-isoprostane were measured using different bio-samples in the relevant RCTs. Four RCTs measured urinary levels (three for MDA [Allen et al., 2011;Cui et al., 2018;Li et al., 2017] and three for 8-isoprostane [Allen et al., 2011;Bräuner et al., 2008;Liu et al., 2020b]), and two RCTs assessed the concentrations in exhaled breath condensate (one for MDA [Dong et al., 2019] and one for 8-isoprostane [Shao et al., 2017]). However, they failed to find any significant associations. Besides, one RCT measured serum-level MDA and 8-isoprostane, reporting significant reductions associated with air purification (Li et al., 2017). Table 2 presents a summary of findings for the certainty of evidence for each health outcome meta-analysed. As many studies were assessed as having "some concerns" or "high" risk of bias based on the RoB2 tool, the risk-of-bias certainty assessment was "serious" for all outcomes. Serious indirectness (significant heterogeneity in population) and imprecision (wide 95% CIs and small sample size) across studies were the main reasons for downgrading the certainty of evidence for each health outcome. Egger's tests and funnel plots found no evidence of publication bias for any health outcomes examined, suggesting unsuspected publication bias. In summary, the certainty of evidence across the indoor air purification RCTs were assessed as "very low" for all health outcomes, implying that more rigorous studies are very likely to change the estimated effects.

Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts

Description for ongoing RCTs
The search in ClinicalTrails.gov identified three relevant ongoing RCTs registered in 2017-2020 (Appendix I in the Supplementary file). All three RCTs employ HEPA air purifiers and recruit middle-aged adults or elderly non-smokers, with one adopted a longer-term (12 months) parallel design examining endothelial function and cognitive impairment, and the rest assessing short-term effects (~30 days) with a crossover design on BP, HRV and biomarkers, but only one of them has planned for a washout period.

Discussion
To the best of our knowledge, this is the first systematic review and meta-analysis to comprehensively and quantitatively evaluate the cardiovascular effects of indoor air purification interventions in RCTs assessing BP, PP, RHI, CRP, IL-6 and fibrinogen. The 14 crossover RCTs involving more than 700 participants from four countries show that air purification interventions consistently lead to a substantial reduction of indoor PM 2.5 levels (mean=56%) and lower SBP (by ~2.5 mmHg). There is also suggestive evidence on reduced PP and increased RHI in RCTs using physical-type air purifiers only. However, the overall certainty of evidence remains low due to a range of study limitations identified, warranting larger and more robust studies to clarify the cardiovascular benefits of air purification interventions.

Blood pressure
Elevated BP is the second leading risk factor of premature death globally and is considered a major pathway linking air pollution to CVD (Newby et al., 2015). In air purification RCTs, BP has been the most widely studied cardiovascular outcome. Compared to an earlier review of nine crossover trials and one before-and-after study (Walzer et al., 2020), we included exclusively RCT evidence (with four more trials included) (Allen et al., 2011;Dong et al., 2019;Liu et al., 2020b;Weichenthal et al., 2013) and excluded an under-powered RCT that reported inappropriate effect estimate for BP (Karottki et al., 2013). Using a more updated risk-of-bias assessment tool tailored for crossover RCTs, we highlighted the key methodological concerns in the literature more clearly than the previous review (3/12 vs 8/10 studies judged as having low risk-of-bias) and concluded a "very low" certainty of evidence on the benefits of air purifier intervention on BP. Furthermore, we found considerably smaller overall reductions in SBP (-2.28 -23.4, -6.85] mmHg) and DBP (-5.00 [-12.54, 2.54] mmHg) that may be biased due to the lack of randomization (Lin et al., 2011). In addition, the previous meta-analysis appeared to have inappropriately combined geometric means with arithmetic means directly (Walzer et al., 2020;Higgins et al. 2008), whereas we have transformed the data into comparable forms for the meta-analysis.
From the subgroup analyses, there is indicative evidence showing a slightly greater reduction in SBP in trials involving individuals with higher baseline SBP (≥120 mmHg), conducted at home (vs. schools), using physical-type air purifiers only, recorded lower Europe PMC Funders Author Manuscripts baseline PM 2.5 levels (≤25 μg/m 3 ), and with lower risk of bias. Although such differences did not reach statistical significance, possibly due to the limited power, they are well expected. First, hypertensive individuals tend to be more vulnerable to the adverse effect of PM 2.5 (Auchincloss et al., 2008). Second, the RCTs conducted in classrooms or dormitories usually set one air purifier in each room for three to eight participants, which might have weakened the effects in reducing personal PM 2.5 exposure. Third, pervious study reported that electrostatic or ionization air purifiers could produce ozone, a highly reactive gas associated with increased cardiovascular risks (Srebot et al., 2009), during their electric charging process (Michael et al., 2008). This may explain the greater effect size observed in the RCTs using physical-type air purifiers. Fourth, given the supra-linear association between PM 2.5 and cardiovascular disease, the same proportional reduction of PM 2.5 levels in low pollution settings should result in greater benefits (Pope III et al., 2018). Fifth, studies with high risk of bias have various issues (e.g. deviation from intended intervention) that might have weakened the interventions. These offer important insight into more efficient design of future RCTs with greater power to detect modest short-term effects.

Other cardiovascular biomarkers
Among other cardiovascular biomarkers, PP, endothelial dysfunction, and inflammatory biomarkers (e.g. CRP, IL-6, fibrinogen) are well-established predictors of cardiovascular disease risk (Bourdrel et al., 2017;Winston et al., 2013). Similar to the conclusion drawn on BP, the existing RCT evidence on the effects of air purifier interventions on the above biomarkers remains inconclusive.
Exposure to PM 2.5 has been shown to be associated with higher PP and HRV and lower endothelial function (Auchincloss et al., 2008;Niu et al., 2020;Schlesinger, 2007). However, our review found no significant benefits in the overall analysis on PP, but the effect estimate became marginally statistically significant when restricting the analysis to RCTs using physical-type air purifiers only. Similarly, a significant improvement was also observed in RHI among the RCTs using physical-type air purifiers only. In addition, only the RCT using ionization air purifier reported significantly lower HRV (Dong et al., 2019), which may reflect an adverse cardiovascular effect of the ozone produced by electrostatic and ionization air purifiers.
Previous in vivo and in vitro experimental studies have consistently shown negative impact of PM 2.5 exposure on systemic inflammation (Münzel et al., 2017), and similar findings have been reported in human experiments on short-term concentrated PM 2.5 exposure challenge (Pope III et al., 2016). Although we found non-significant changes in CRP, IL-6 and fibrinogen (biomarkers of inflammation) after indoor air filtration interventions, it should be noted that this is based on the limited number of studies identified and there are moderate to high risk of bias in most RCTs. In addition to the lack of power, the short duration and variability in PM 2.5 reduction of intervention might have resulted in modest effects that can be easily masked by noise.
A wide range of other cardiovascular biomarkers (e.g. MDA, 8-isoprostane) have been investigated previously, but the evidence remains scattered and no reliable meta-analysis can be performed. Although some of the signals could serve as indications for further investigation, no reliable conclusion can be drawn from the totality of evidence given the small sample sizes and lack of consistency across studies. Considering the overall quality of the evidence on the more frequently investigated health outcomes, interpretation on the less well-studied biomarkers must be even more cautious.

Methodological limitations and knowledge gaps
In this review, we identified a range of methodological limitations in the previous RCTs.
First, most previous studies were shot-term interventions in small samples, which may underestimate the benefits of reducing PM 2.5 exposure. Notably, the largest (n=200) and longest (intervention duration=1 year) RCT (with little risk of bias) reported highly significant effects on SBP (-7.70 [-10.0, -5.40] mmHg), DBP (-3.20 [-4.50, -1.90] mmHg), and CRP (-2.80 [-2.98, -2.62] mg/L) of clinical relevance (Chuang et al., 2017), but these findings are masked in the random-effect models that assigned disproportionately smaller weight to this study. Second, only half of the trials incorporated a wash-out period (and most were relatively short, median=7 days), which is crucial to minimise carry-over effects, a major challenge in crossover RCTs that typically bias the effect estimate towards the null. Third, the included studies were highly heterogeneous. Subgroup analyses were conducted to investigate this, but there was limited power to detect true subgroup-difference. In particular, the report of effect estimates other than mean differences (e.g. median) without supplementary information (e.g. inter-quartile range) in some studies may have introduced additional noise to our meta-analyses, because indirect approximation of means and SDs were required that entails extra uncertainty. In particular, the inconsistent reporting of percentage change in arithmetic and geometric mean across intervention arms without providing the exact group-specific means prevented us from conducting meta-analysis on some biomarkers. Besides, the indirect approximation of means and SDs in some studies might have introduced additional noise to our meta-analyses. Fourth, since most of the studies included were either conducted in China or a primarily-Caucasian population (i.e. Denmark, USA, Canada), there is a lack of evidence from more diverse population with different air pollution exposure patterns. More studies in different populations are needed to assess the potential cardiovascular benefits of air purifiers.
It should also be noted that most of the existing short-term RCTs explored the cardiovascular benefits of using indoor air purifiers under an experimental environment (e.g., by asking the participants to stay in the air-filtered areas as long as possible), which restricted the daily activity patterns of the participants. Therefore, in the real-world settings where people would spend less time in the air-filtered areas, the cardiovascular benefits of using indoor air purifiers might be lower than the estimated effects reported in this review.

Conclusions
Based on 18 articles from 14 independent RCTs, this review suggests statistically significant reduction in SBP (although being modest in absolute term) following indoor air purification interventions, with hints of stronger effect from more robust intervention. There is also indicative evidence of reduced PP and increased RHI, particularly in the RCTs using physical-type air purifiers. In contrast, we found no clear changes in levels of CRP, IL-6 and Europe PMC Funders Author Manuscripts fibrinogen, and there is even less evidence on other cardiovascular biomarkers. According to the Cochrane RoB2 criteria, most published trials suffered from moderate to high risk of bias, contributing to the "very low" overall certainty of evidence under the GRADE framework. Besides, there are a range of methodological limitations in the existing RCTs, particularly small sample size, short intervention duration, and lack of wash-out period.
Future high-quality studies with larger sample size, longer intervention period, more robust medium-to-long term endpoints (e.g. carotid intima-media thickness) are urgently needed to clarify the cardiovascular benefits of air purifier interventions, and claims on such benefits should be more cautious before more conclusive evidence.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material. Air purifier usage was significantly associated with an decrease in blood pressure.