Open Air Scholar

Design:

A systematic review and meta-analysis of prospective cohort studies published up to April 2024. Studies were included if they used multivariate regression to assess risk factors for POD in adult cardiac surgery patients and employed validated delirium assessment tools (e.g., CAM-ICU, DSM-IV, ICDSC). Excluded were univariate analyses and non-prospective designs.

Procedures:

Databases (Scopus, PubMed, Web of Science) were searched using keywords targeting cardiac surgery types, delirium assessment tools, and risk factors. Risk of bias was evaluated using the Newcastle-Ottawa Scale (NOS), with 17 studies rated 'good' (≥7 stars) and six 'fair.' Statistical analyses included pooled odds ratios (ORs) with 95% confidence intervals (CIs), using fixed- or random-effects models based on heterogeneity (I² >50% or p <0.1). Subgroup analyses explored sources of heterogeneity for age, depression, and postoperative AF. Sensitivity analyses excluded individual studies to test robustness, and publication bias was assessed via funnel plots and Egger’s test.

Primary Findings:

Eight risk factors were significantly associated with POD: age (per year increase: OR 1.06, 95% CI 1.04–1.08, p <0.001), pre-operative depression (OR 3.71, 95% CI 2.45–5.62, p <0.001), post-operative atrial fibrillation (AF: OR 2.39, 95% CI 1.79–3.21, p <0.001), age ≥65 years (OR 3.32, 95% CI 2.40–4.60, p <0.001), pre-operative AF (OR 4.43, 95% CI 2.56–7.69, p <0.001), diabetes mellitus (OR 2.16, 95% CI 1.39–3.35, p =0.001), combined CABG + valve surgery (OR 2.73, 95% CI 1.66–4.49, p <0.001), and CPB time (per minute: OR 1.02, 95% CI 1.01–1.04, p =0.001). Hypertension showed no significant association (OR 1.64, 95% CI 0.75–3.56, p =0.212).

Secondary Findings:

Heterogeneity was moderate to high for age (I² =57.4%), CPB time (I² =57.2%), and hypertension (I² =84.9%). Subgroup analyses revealed stronger age effects in smaller studies (<250 patients) and higher risks of depression in younger populations. Postoperative AF effects were consistent across subgroups. Publication bias was detected for postoperative AF (p=0.018), but not for age or depression. Sensitivity analyses confirmed the robustness of age’s association with POD.

Interpretation:

The findings highlight aging’s cumulative impact on cerebral reserve and neuroinflammatory pathways as key mechanisms for POD. Pre-operative depression and diabetes may exacerbate delirium via neurotransmitter dysregulation and blood-brain barrier disruption. Surgical complexity (e.g., CABG + valve) and prolonged CPB time likely reflect increased perioperative stress and cerebral hypoperfusion. The study expands on prior reviews by identifying pre-operative AF as a critical risk factor and emphasizing the role of depression screening. However, conflicting results on hypertension underscore the need for standardized definitions and larger cohorts.

Strengths:

Comprehensive search strategy, inclusion of high-quality studies (NOS ≥7 stars), and focus on multivariate analyses minimized confounding. Subgroup and sensitivity analyses clarified heterogeneity sources, enhancing clinical relevance.

Limitations:

Exclusion of non-prospective studies may have omitted valuable data. Heterogeneity in hypertension and CPB time suggests variability in surgical practices or patient populations. Diagnostic tool variability (e.g., CAM-ICU vs. ICDSC) and publication bias in AF outcomes limit generalizability.

Design:

The study employed a combination of structural and biochemical approaches to characterize the terminase complex. Researchers co-expressed gp3 and gp2 in E. coli, purified the complex via amylose affinity chromatography and gel filtration, and analyzed it using negative-stain transmission electron microscopy (TEM), native matrix-assisted laser desorption/ionization (MALDI) mass spectrometry (MS), and native agarose gel electrophoresis. Crosslinking experiments with glutaraldehyde were used to probe interactions between terminase and empty viral procapsids.

Procedures:

TEM utilized a Random Conical Tilt (RCT) strategy to capture 3D structural heterogeneity, processing over 35,000 particles with iterative Stable Alignment and Classification (ISAC) to identify distinct assembly states. Native MS measured molecular weights of terminase complexes, while native agarose gels assessed terminase-procapsid binding stability. Crosslinking experiments tested glutaraldehyde concentrations (0.01% and 0.1%) to stabilize transient interactions. Cryo-EM and orthogonal tilt data acquisition were also applied to refine structural analyses.

Primary Findings:

TEM revealed three major terminase configurations: (gp3)₉:2(gp2), (gp3)₉:5(gp2), and (gp3)₉:7(gp2). The gp3 subunit formed a stable nonamer ring, consistent with crystallographic data, while gp2 assembled into ellipsoid oligomers. Native MS confirmed these stoichiometries, with molecular weights of ~290 kDa (1(gp3)₉:2(gp2)), ~642 kDa (2(gp3)₉:5(gp2)), and ~925 kDa (3(gp3)₉:7(gp2)). Crosslinking experiments showed that terminase binds procapsids transiently, forming stable complexes only when crosslinked at low glutaraldehyde concentrations (0.01%). Higher concentrations induced aggregation.

Secondary Findings:

The (gp3)₉:5(gp2) complex was proposed as the physiological form for DNA packaging, aligning with pentameric gp2 structures observed in other phages. The terminase's structural flexibility was linked to disordered regions in gp3 and gp2, enabling dynamic interactions with the viral capsid. ISAC classification identified distinct populations of terminase complexes, including forms with multiple gp3 nonamers attached to gp2 cores.

Interpretation:

The terminase's heterogeneity suggests a dynamic equilibrium between assembly states, enabling adaptation to the portal protein's 5-fold symmetry and precise DNA packaging. The transient terminase-procapsid interaction implies a role in initiating DNA translocation, after which the complex dissociates to allow tail assembly. The findings parallel observations in herpesviruses, where terminase flexibility correlates with functional diversity. The gp2 oligomeric states may regulate nuclease activity timing, ensuring genome-length precision.

Strengths:

The study combined multiple techniques (TEM, MS, electrophoresis) to corroborate findings, providing a holistic view of terminase dynamics. The use of ISAC and RCT advanced structural analysis of heterogeneous complexes.

Limitations:

High-resolution cryo-EM structures were not obtained due to terminase instability outside the viral context. Future studies may require crosslinking or DNA supplementation to stabilize complexes.

Design:

A prospective observational study enrolled 51 consecutive RCC patients from a specialist clinic. Participants underwent genotyping for RFC1 expansions, clinical assessments, and specialized evaluations for those with RFC1++.

Procedures:

Demographics, cough severity (visual analog scale [VAS], Leicester Cough Questionnaire [LCQ]), and spirometry were assessed in all patients. RFC1++ carriers underwent nerve conduction studies, brain MRI, and capsaicin cough challenge tests to measure reflex sensitivity. Statistical analyses compared clinical and physiological data between RFC1++ (n=4), carriers (RFC1+−, n=5), and wild-type (RFC1−−, n=42) groups.

Primary Findings:

RFC1++ was identified in 8% (4/51) of patients, with 10% carriers and 82% wild-type. No significant differences were observed in age, sex, cough duration, spirometry, or LCQ scores between RFC1++ and RFC1−− groups. However, pins and needles symptoms were more frequent in RFC1++ (100% vs. 33% in RFC1−−, p=0.01). Nerve conduction studies revealed sensory neuropathy in all RFC1++ patients, and one exhibited cerebellar atrophy on MRI. Cough reflex sensitivity (capsaicin challenge) was heightened in RFC1++ (C2: 1.1±4.4 μmol/L⁻¹; C5: 1.9±3.6 μmol/L⁻¹), comparable to historical RCC data but far lower than healthy controls.

Secondary Findings:

RFC1++ patients showed no significant differences in cough severity (VAS/LCQ) or spirometry compared to non-carriers, suggesting overlapping clinical presentations. The cerebellar atrophy in one RFC1++ patient aligns with CANVAS neuropathology, though neurological symptoms were not universally severe.

Interpretation:

RFC1++ accounts for 8% of RCC cases in this cohort, highlighting its role as a significant but underrecognized cause. While clinical features overlap with non-genetic RCC, sensory neuropathy and pins and needles symptoms may aid identification. The heightened cough reflex sensitivity in RFC1++ supports neural hypersensitivity mechanisms, consistent with CANVAS pathophysiology. However, the lack of severe neurological deficits in most patients suggests a spectrum of RFC1-related disease.

Strengths:

The study’s specialist cohort and comprehensive evaluations (nerve conduction, MRI, capsaicin testing) provide robust data on RFC1++ in RCC. Identification of pins and needles as a distinguishing feature offers a practical clinical clue.

Limitations:

Small sample size limits statistical power, particularly for subgroup comparisons. Nerve conduction and MRI data were not collected for non-RFC1++ patients, complicating direct comparisons. Cross-sectional design precludes longitudinal insights into symptom progression.

Design:

The study employed a multi-step synthesis and validation process. FeNPs@NCNT was synthesized via high-temperature pyrolysis of metal-organic frameworks (MOFs), specifically ZIF-8 and iron-containing MIL-101, followed by nitrogen doping. The nanozyme’s oxidase and peroxidase-like activities were characterized using electron microscopy, X-ray diffraction (XRD), and X-ray photoelectron spectroscopy (XPS). TAC assays were performed using OPD oxidation, with quercetin as a model antioxidant. The method was validated against the standard ABTS assay using plant extracts and commercial beverages.

Procedures:

FeNPs@NCNT synthesis involved ball-milling MOF precursors, pyrolysis at 900°C under nitrogen, and ammonia treatment for nitrogen doping. Structural characterization confirmed bamboo-like carbon nanotubes with 5–10 nm iron nanoparticles and nitrogen-doped carbon. TAC assays measured OPD oxidation inhibition by antioxidants, with colorimetric analysis via a smartphone’s RGB sensor (Color Picker APP) and fluorescence spectroscopy (excitation/emission: 417/568 nm). Reaction conditions were optimized for FeNPs@NCNT concentration (30 μg/mL), pH (6.5–7.2), and reaction time (30 minutes. Selectivity was tested against ions (MgCl₂, NaNO₂) and biomolecules (BSA, glucose). Stability was assessed over five weeks.

Primary Findings:

FeNPs@NCNT exhibited oxidase-like activity, catalyzing OPD oxidation to DAP via superoxide anion (O₂⁻) generation. The smartphone RGB method showed a linear response to quercetin (0.1–15 μg/mL, R² = 0.9964) with a LOD of 0.06 μg/mL. Fluorescence detection extended sensitivity to 3.1 ng/mL (linear range 50–20,000 ng/mL. The nanozyme’s Michaelis-Menten parameters (Km = 0.042 mM, vmax = 1.76×10⁻⁸ M/s) indicated high catalytic efficiency. Real-world validation with plant extracts (e.g., Carthami flos: 12.8 μmol TE/g) and beverages showed strong correlation (R² > 0.95) with ABTS results.

Secondary Findings:

SEM/TEM revealed uniformly distributed FeNPs (5–10 nm) within carbon nanotubes, while XPS confirmed nitrogen doping (pyridinic, pyrrolic, and oxidized species) and metallic Fe³⁺ phases. EPR confirmed O₂⁻ and ·OH radical formation during catalysis. The method showed minimal interference from common ions (e.g., MgCl₂, NaNO₂) but relied on quercetin’s unique ROS scavenging. The nanozyme retained >90% activity after five weeks, demonstrating practical stability.

Interpretation:

The nanozyme’s structure—FeNPs within nitrogen-doped carbon—enhances catalytic efficiency through synergistic effects of metallic iron and nitrogen functional groups. Eliminating H₂O₂ simplifies protocols and avoidses its instability, broadening applicability. The smartphone-based RGB method enables field use in resource-limited settings, while fluorescence offers lab-grade sensitivity. The strong correlation with ABTS validates its accuracy for real-world samples. The oxidase mode’s oxygen dependency confirms the mechanism relies on ambient O₂, avoiding external oxidants.

Strengths:

The method combines high sensitivity (LOD down to 3.1 ng/mL), portability (smartphone integration), and eco-friendly synthesis from MOFs. The dual detection modes (RGB/fluorescence) provide flexibility for different settings. Validation across diverse samples (eight plant extracts, beverages) and strong ABTS correlation underscore robustness.

Limitations:

While selective against common interferents, the method’s performance in highly complex matrices (e.g., high-protein foods) requires further testing. Long-term stability beyond five weeks and scalability for industrial use remain unexplored. Fluorescence requires specialized equipment, limiting its field applicability compared to the smartphone method.

Design:

The study enrolled 159 hospitalized COVID-19 survivors and 29 age-, sex-, and cardiovascular risk-matched controls. Participants underwent computed tomography pulmonary angiography (CTPA) and high-resolution computed tomography (HRCT) imaging 28–60 days post-discharge. Blood biomarker analysis and patient-reported outcomes (e.g., EQ-5D-5L, Duke Activity Status Index) were also collected. Logistic regression models were used to identify predictors of persistent radiological abnormalities, defined as ground-glass opacity (GGO) and/or reticular changes.

Procedures:

Key procedures included multimodal imaging to assess lung parenchyma, measurement of acute-phase biomarkers (e.g., D-dimer, ferritin) during hospitalization, and convalescent biomarkers (e.g., ICAM-1, VCAM-1, endothelin-1). Patient-reported outcomes assessed physical activity, quality of life, and functional capacity (e.g., predicted maximal oxygen utilisation [VO2]). Statistical analyses adjusted for age, sex, and comorbidities, with subgroup comparisons between patients and controls.

Primary Findings:

Fifty-six percent of patients retained radiological abnormalities at follow-up, including GGO (45%), reticulation/architectural distortion (30%), or mixed patterns (19%), which were rare in controls. Persistent abnormalities correlated with older age, worse pre-COVID performance status, severe acute illness markers (e.g., hypoxia, elevated D-dimer), and elevated convalescent biomarkers like ICAM-1 (OR 6.23) and VCAM-1 (OR 5.46). Clinical factors such as non-invasive ventilation (OR 31.09) and prophylactic anticoagulation (OR 7.26) also predicted persistence. Patients with abnormalities reported lower physical activity, reduced VO2, and poorer EQ-5D-5L scores.

Secondary Findings:

Most patients (68%) showed improvement from admission imaging, with <20% of lung parenchyma remaining abnormal. Elevated endothelin-1 (OR 3.34) and LDH (OR 3.28) at follow-up further predicted radiological persistence. Biomarkers like ferritin (OR 6.82) and peak D-dimer (OR 2.57) during hospitalization were strongly associated with persistent changes. No significant differences in lung function tests (e.g., diffusing capacity) were reported due to methodological limitations.

Interpretation:

Persistent radiological abnormalities reflect unresolved pulmonary inflammation or fibrosis, linked to impaired physical function and quality of life. The findings underscore the role of endothelial dysfunction (marked by ICAM-1 and VCAM-1) and hypercoagulability (elevated D-dimer) in post-acute lung damage. The association with anticoagulation suggests complex pathophysiological mechanisms requiring further study. These results align with prior evidence on post-viral fibrosis and highlight the need for early intervention to mitigate long-term sequelae.

Strengths:

The study combined multimodal imaging with longitudinal clinical and biomarker data, enhancing the robustness of associations. The inclusion of matched controls and patient-reported outcomes provided a holistic view of post-acute impacts. Multivariable analyses adjusted for confounders, strengthening causal inferences.

Limitations:

The observational design precluded causal conclusions. Reliance on self-reported outcomes and lack of lung function tests (e.g., DLCO) limited the assessment of physiological impairment. The 60-day follow-up may not capture long-term trajectories, and the sample size restricted subgroup analyses.

Design:

The study employed a cross-sectional design with a retrospective review of medical records from SMI inpatients admitted to VBH during the pandemic. Participants included 254 individuals, 75% male, with a mean age of 35.7 years. The primary psychiatric diagnoses were schizophrenia (37%), bipolar disorder (21%), and schizoaffective disorder (19%). Comorbidities such as hypertension (11%), HIV (7%), and nicotine use (71%) were documented. The cohort was limited to patients deemed medically stable upon admission, reflecting VBH’s standard criteria for accepting psychiatric patients. This selection may have introduced bias toward milder cases.

Procedures:

Data collection relied on standardized tools: the VBH COVID-19 Tracker and Patient Screening Control Form. The Tracker monitored symptoms (e.g., fever, cough), vital signs (heart rate, oxygen saturation), psychiatric status, and comorbidities during hospitalization. The screening form assessed exposure risks, travel history, and immediate health indicators upon admission. Statistical analyses identified associations between patient characteristics, symptom severity, and outcomes such as transfer to a medical ward or mortality. The study adhered to South African Good Clinical Practice guidelines and underwent ethical review, with no declared competing interests.

Primary Findings:

Of the 254 patients, 62% exhibited symptoms of COVID-19, with 7% requiring transfer to a medical ward and 1% (2 patients) dying. Both fatalities involved individuals over 40 years old residing in long-stay wards, who had multiple comorbidities such as hypertension and HIV. Transferred patients were significantly older (mean age 42 vs. 35 years) and presented with abnormal vital signs, including elevated heart rates, respiratory distress, and low oxygen saturation (≤92%). These findings underscored age and physiological instability as critical risk factors for severe outcomes.

Secondary Findings:

The 'other comorbidities' category (e.g., diabetes, renal disease) showed a statistically significant association with transfer to medical care. A history of pulmonary tuberculosis (PTB) approached significance (p=0.06), suggesting a potential link between prior TB and severe illness. Notably, despite high nicotine use (71%), no significant correlation was found between substance use and adverse outcomes. The majority (99%) of patients recovered, aligning with global reports of lower mortality in younger populations with fewer comorbidities. The study also highlighted the importance of respiratory rate and oxygen levels as transfer criteria, as these vital signs were strongly associated with medical referral.

Interpretation:

The findings revealed that most SMI patients at VBH experienced mild COVID-19, contrasting with early pandemic predictions of high mortality. Protective factors included the cohort’s younger age and lower prevalence of severe comorbidities compared to the general population. However, older patients and those with abnormal vital signs or multimorbidity faced elevated risks, consistent with global evidence. The near-significant association with PTB warrants further investigation, given South Africa’s high TB burden. The study’s use of structured screening tools demonstrated their utility in managing outbreaks, though their effectiveness in diverse or larger populations remains untested. The results emphasize the need for tailored protocols in psychiatric settings to balance infection control with mental health care.

Strengths:

Key strengths included the use of standardized tools to track symptoms and comorbidities, providing actionable insights for clinical management. The real-world data from a specialized psychiatric hospital offered context-specific evidence, addressing gaps in understanding SMI populations’ pandemic resilience. The study also contextualized outcomes within broader discussions of multimorbidity and institutional risks, reinforcing the importance of integrated care models.

Limitations:

Limitations arose from the retrospective design, which may have introduced recall bias or incomplete data capture. The exclusion of medically unstable patients likely skewed results toward milder cases, limiting generalizability. Small sample size restricted subgroup analyses, particularly for less common diagnoses or comorbidities. Additionally, clinician-reported admission data might have underrepresented certain conditions, such as undiagnosed hypertension or HIV.

Design:

A cross-sectional study enrolled 20 PDAC patients, 10 DC/AC patients, and 20 HCs, matched for age, gender, and BMI. Saliva samples were collected after fasting and avoiding oral hygiene, while pancreatic tissues were obtained during pancreaticoduodenectomy under sterile conditions. Environmental controls included saline samples from surgical and collection rooms. DNA extraction, 16S rRNA V3-V4 region sequencing, and bioinformatics analysis (Kraken2) were used to profile microbial communities. Statistical analyses compared alpha diversity (Shannon, Simpson indices) and taxonomic composition.

Procedures:

Saliva and pancreatic tissue samples were processed using standardized protocols. DNA extraction followed by PCR amplification targeted the V3-V4 hypervariable regions of the 16S rRNA gene. Taxonomic profiling identified microbial phyla, genera, and species. Environmental contamination was assessed via saline controls to distinguish true microbiota from room-derived microbes. Statistical comparisons included ANOVA and PERMANOVA to evaluate group differences.

Primary Findings:

Saliva from PDAC and DC/AC patients showed significantly lower microbial diversity (Shannon index) than HCs, with no difference between cancer groups. Pancreatic tissues of PDAC patients exhibited higher diversity than DC/AC, dominated by Firmicutes (75%), Proteobacteria, and Actinobacteria. Oral-derived genera like Clostridium perfringens and Anoxybacillus were enriched in PDAC tissues, suggesting translocation. Environmental saline samples contained genera (e.g., Anoxybacillus, Halomonas) overlapping with patient samples, necessitating contamination controls. Key species in PDAC tissues included C. perfringens, correlating with oral microbiota.

Secondary Findings:

HC saliva was enriched in Streptococcus, while cancer patients had higher Prevotella and Rothia. PDAC tissues showed abundant Clostridium and Bacillus, whereas DC/AC tissues had Halomonas and Sphingomonas. Shared genera (e.g., Streptococcus, Clostridium) between saliva and pancreatic tissues in PDAC patients supported translocation hypotheses. Environmental controls revealed low but detectable microbial abundance, highlighting contamination risks in microbiome studies.

Interpretation:

Oral dysbiosis in cancer patients may reflect systemic microbial shifts, with pathogenic taxa like Prevotella potentially contributing to tumor progression. Translocation of oral bacteria (e.g., C. perfringens) to pancreatic tissues suggests tumor microenvironments, such as lactate-rich niches, facilitate colonization. Higher diversity in PDAC tissues may indicate immune infiltration or metabolic interactions. Environmental contamination underscores the need for rigorous controls to avoid false positives. The findings suggest microbial signatures could serve as diagnostic or prognostic markers.

Strengths:

First to compare pancreatic microbiota between PDAC and DC/AC, incorporating environmental controls. Standardized protocols minimized technical biases, and species-level analysis provided mechanistic insights. The matched design strengthened translocation hypotheses.

Limitations:

Reliance on surgical samples limited generalizability. Healthy pancreatic tissue controls were unavailable, complicating comparisons. Cross-sectional design precluded longitudinal insights into microbiome dynamics. Contamination risks, though addressed, remain a challenge in sterile tissue studies.

Design:

The study employed a prospective cohort design using data from the JECS, which enrolled 90,693 mother-infant pairs after excluding multiple births, miscarriages, and incomplete records. Participants were followed from early pregnancy through childbirth, with outcomes derived from medical records. Air purifier use was self-reported during pregnancy, and ABOs were defined using standardized criteria: PTB (<37 weeks), SGA (birth weight <10th percentile for gestational age), and LBW (<2,500g at birth).

Procedures:

Logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for associations between air purifier use and ABOs. Adjustments were made for demographic factors (e.g., maternal age, education), socioeconomic status, environmental exposures (e.g., smoking, traffic proximity), and pre-pregnancy health conditions. Subgroup analyses examined sex-specific effects and timing of LBW reduction in full-term births. Multiple imputation addressed missing data, and sensitivity analyses tested robustness of findings.

Primary Findings:

Crude analyses showed reduced risks of PTB (OR 0.92), SGA (OR 0.94), and LBW (OR 0.89) with air purifier use. After adjusting for confounders, significant associations remained only for SGA (OR 0.94, 95% CI 0.89–1.00, p=0.048) and LBW (OR 0.93, 95% CI 0.88–0.98, p=0.003). PTB was not significantly associated with purifier use in adjusted models. LBW risk reduction was specific to male infants (adjusted OR 0.91, 95% CI 0.85–0.98, p=0.017), while no association was found with congenital anomalies.

Secondary Findings:

LBW reduction persisted in full-term births (adjusted OR 0.94, 95% CI 0.89–0.99), suggesting effects on fetal growth rather than gestational duration. No significant sex difference was observed for SGA, though reduced statistical power due to stratification may have obscured biological effects. Sensitivity analyses confirmed robustness of findings across different model specifications.

Interpretation:

The findings suggest air purifiers may reduce pollution-related risks of SGA and LBW, particularly in male infants, potentially due to sex-specific fetal vulnerability to oxidative stress and inflammation caused by pollutants like PM2.5. The lack of association with PTB aligns with prior evidence that PM2.5 exposure is more consistently linked to fetal growth restriction than preterm labor. Mechanisms may involve reduced placental inflammation and improved nutrient transport, though direct evidence requires further study.

Strengths:

The study’s strengths include its large sample size (n=90,693), prospective design minimizing recall bias, and rigorous adjustment for confounders using JECS’s comprehensive data on maternal health, lifestyle, and environmental exposures. The nationally representative cohort enhances generalizability to Japanese populations.

Limitations:

Key limitations include reliance on self-reported air purifier use without data on device efficacy, filter type, or usage duration. Indoor air quality measurements were unavailable, precluding mediation analysis. Residual confounding from unmeasured factors (e.g., paternal exposure, dietary habits) remains possible. Additionally, subgroup analyses may have reduced statistical power.

Design:

The study employs a causal-inference machine learning framework to disentangle the effects of meteorological and anthropogenic factors on ozone levels. It combines observational data from ground-based air quality monitoring stations, satellite-derived atmospheric precursors (e.g., formaldehyde and NO₂ from TROPOMI), and meteorological variables from ERA5 reanalysis. A spatially comprehensive analysis covers China’s major regions, including the Yangtze River Delta (YRD), Beijing-Tianjin-Hebei (BTH), and Sichuan Basin (SCB).

Procedures:

Key methodologies include: (1) Random Forest (RF) models to quantify variable importance in ozone formation; (2) SHAP analysis to interpret individual feature contributions; (3) Causal Forest (CF) models to estimate Average Treatment Effects (ATE) for meteorological drivers. The study also analyzes the HCHO-to-NO₂ ratio to assess shifts in ozone formation regimes. Satellite and ground data are integrated to validate findings and account for spatial variability.

Primary Findings:

Meteorological factors, particularly temperature (T2m) and solar radiation (SSRD), were the dominant drivers of the 2022 ozone rebound, collectively explaining 57% of the increase. SHAP analysis revealed temperature contributed 38.9%, solar radiation 18.47%, and other meteorological variables (relative humidity, boundary layer height, wind speed) an additional 8.55%. Anthropogenic factors like reduced PM₂.₅ and NOx emissions dampened ozone levels but were outweighed by meteorological influences. The rebound was most pronounced in the YRD, BTH, and SCB regions, where temperature rose by 1.64°C and solar radiation increased by 5.12% compared to previous years.

Secondary Findings:

A 90% increase in the HCHO-to-NO₂ ratio at 90% of monitoring stations indicated a shift toward VOC-limited or NOx-limited ozone regimes, driven by reduced NOx emissions and rising temperatures. Seasonal analysis showed ozone increases in all seasons except winter, with summer exhibiting the strongest meteorological influence. Causal forest models confirmed temperature as the primary causal driver, particularly in central and eastern China, while wind speed effects varied regionally. Formaldehyde levels rose by 6.32%, and PM₂.₅ declined by 14.9 mg/m³ since 2013, underscoring the complexity of pollutant interactions.

Interpretation:

The findings emphasize that climate-driven meteorological shifts, such as rising temperatures and solar radiation, are critical to ozone dynamics. Temperature’s dominant role reflects its dual impact on chemical reaction rates and biogenic VOC emissions. The observed regime shifts in ozone formation highlight the need for region-specific emission controls. For instance, NOx reductions in VOC-limited areas may exacerbate ozone, while VOC management remains critical in NOx-limited regions. The study underscores that emission controls alone may not mitigate ozone pollution without addressing climate-driven meteorological changes.

Strengths:

The integration of causal inference with machine learning allowed robust disentanglement of confounding variables, offering a methodological advancement over traditional statistical approaches. Satellite data (e.g., TROPOMI) enhanced spatial coverage and validation of ground observations. The analysis of HCHO-NO₂ ratios provided actionable insights into regional ozone regimes.

Limitations:

Causal forest models rely on linear approximations, potentially underestimating nonlinear meteorological effects. Regional variations in emission inventories and biogenic sources may introduce uncertainties. Additionally, the study focuses on 2022, limiting insights into long-term climate trends.