Regional effects unlikely to explain association between ozone and cardiovascular mortality in China – Authors' reply

Abstract

We thank Kouji H Harada and Mariko Harada Sassa for their comments on our recent Article on long-term ozone exposure and cardiovascular mortality.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar We hereby provide additional information on our findings in response to the issue they raiased.We agree with Harada and Harada Sassa that the stratum-specific estimates should be compared with the overall estimate. However, Harada and Harada Sassa seemed to confuse confounding with effect modification. Therefore, here we provide two decision trees to simply distinguish between confounding and effect modification (appendix pp 1–2). As pointed out by Harada and Harada Sassa, if there is no interaction or effect modification, the stratum-specific estimates should be comparable and the average should be similar to the overall estimate.As for confounding, there are two common and simple ways to determine its presence. One is to fit a model adjusted for the potential confounder and then compare the unadjusted and adjusted estimates (appendix p 1). The other is to compare the stratum-specific estimates with the unadjusted overall estimate (appendix p 2), rather than comparing the stratum-specific estimates with the adjusted overall estimate, as Harada and Harada Sassa did. For the first method, confounding might exist if the unadjusted estimate significantly differs from the adjusted estimate. For the second method, confounding might exist if the stratum-specific estimates are both significantly smaller or larger than the unadjusted overall estimate.Following the method described here, we used model 4 in our publication1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar as the adjusted model and removed residence and region from the model, separately, as the corresponding unadjusted models. We then compared (1) the differences between strata by residence or region, (2) the differences between the unadjusted and adjusted estimates, and (3) the differences between stratum-specific estimates and the unadjusted overall estimate.As shown in the appendix (p 3), the effect on cardiovascular mortality was slightly larger in rural areas than in urban areas and in southern China than in northern China, but the between-group differences did not reach significance. We have been fully transparent about these results in our Article.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google ScholarIn additional analysis, we did not find significant differences in the cardiovascular effects of ozone with and without adjustment for residence (p=0·79) or region (p=0·27; appendix p 3). Additionally, estimates of the associations between ozone and cardiovascular mortality in rural areas or in southern China were similar to their corresponding unadjusted overall estimates (p=0·61 for rural areas and p=0·93 for southern China; appendix p 3). Notably, the estimate in southern China lost statistical significance with a wider 95% CI, probably due to the sample size being reduced by nearly half compared with the full study sample. However, we observed that the effects of ozone in urban areas and northern China were relatively smaller than their corresponding unadjusted overall estimate (p=0·066 for urban areas and p=0·028 for northern China). Statistically, the difference between unadjusted overall estimates and the estimate for urban areas or northern China might be partly explained by the differences in estimates between the two strata (ie, urban vs rural or northern vs southern China). Collectively, all these results suggest that the possibility of confounding by residence and region is low but cannot be completely excluded.In our Article,1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar we used Cox proportional hazards models with adjustments for a set of individual-level and regional-level covariates, including residence and region. Fitting multivariate models is a common practice to control for multiple confounders simultaneously.2Pourhoseingholi MA Baghestani AR Vahedi M How to control confounding effects by statistical analysis.Gastroenterol Hepatol Bed Bench. 2012; 5: 79-83PubMed Google Scholar Therefore, we believe that the applied method is appropriate and could statistically exclude the confounding effects of residence, region, and unmeasured variables related to them.Consistent with our findings, Liu and colleagues3Liu S Zhang Y Ma R et al.Long-term exposure to ozone and cardiovascular mortality in a large Chinese cohort.Environ Int. 2022; 165107280Crossref Scopus (1) Google Scholar also observed an independent association between long-term exposure to ozone and cardiovascular mortality in China. Moreover, previous studies have suggested that the associations between ozone exposure and cardiovascular mortality is biologically plausible. For example, ozone inhalation might lead to a systemic inflammatory response, increased release of biomarkers related to coagulation, platelet dysfunction, and endothelial injury, and modulation of the autonomic nervous system and the neuroendocrine system. We have provided these discussions in our Article.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar We thank Kouji H Harada and Mariko Harada Sassa for their comments on our recent Article on long-term ozone exposure and cardiovascular mortality.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar We hereby provide additional information on our findings in response to the issue they raiased. We agree with Harada and Harada Sassa that the stratum-specific estimates should be compared with the overall estimate. However, Harada and Harada Sassa seemed to confuse confounding with effect modification. Therefore, here we provide two decision trees to simply distinguish between confounding and effect modification (appendix pp 1–2). As pointed out by Harada and Harada Sassa, if there is no interaction or effect modification, the stratum-specific estimates should be comparable and the average should be similar to the overall estimate. As for confounding, there are two common and simple ways to determine its presence. One is to fit a model adjusted for the potential confounder and then compare the unadjusted and adjusted estimates (appendix p 1). The other is to compare the stratum-specific estimates with the unadjusted overall estimate (appendix p 2), rather than comparing the stratum-specific estimates with the adjusted overall estimate, as Harada and Harada Sassa did. For the first method, confounding might exist if the unadjusted estimate significantly differs from the adjusted estimate. For the second method, confounding might exist if the stratum-specific estimates are both significantly smaller or larger than the unadjusted overall estimate. Following the method described here, we used model 4 in our publication1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar as the adjusted model and removed residence and region from the model, separately, as the corresponding unadjusted models. We then compared (1) the differences between strata by residence or region, (2) the differences between the unadjusted and adjusted estimates, and (3) the differences between stratum-specific estimates and the unadjusted overall estimate. As shown in the appendix (p 3), the effect on cardiovascular mortality was slightly larger in rural areas than in urban areas and in southern China than in northern China, but the between-group differences did not reach significance. We have been fully transparent about these results in our Article.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar In additional analysis, we did not find significant differences in the cardiovascular effects of ozone with and without adjustment for residence (p=0·79) or region (p=0·27; appendix p 3). Additionally, estimates of the associations between ozone and cardiovascular mortality in rural areas or in southern China were similar to their corresponding unadjusted overall estimates (p=0·61 for rural areas and p=0·93 for southern China; appendix p 3). Notably, the estimate in southern China lost statistical significance with a wider 95% CI, probably due to the sample size being reduced by nearly half compared with the full study sample. However, we observed that the effects of ozone in urban areas and northern China were relatively smaller than their corresponding unadjusted overall estimate (p=0·066 for urban areas and p=0·028 for northern China). Statistically, the difference between unadjusted overall estimates and the estimate for urban areas or northern China might be partly explained by the differences in estimates between the two strata (ie, urban vs rural or northern vs southern China). Collectively, all these results suggest that the possibility of confounding by residence and region is low but cannot be completely excluded. In our Article,1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar we used Cox proportional hazards models with adjustments for a set of individual-level and regional-level covariates, including residence and region. Fitting multivariate models is a common practice to control for multiple confounders simultaneously.2Pourhoseingholi MA Baghestani AR Vahedi M How to control confounding effects by statistical analysis.Gastroenterol Hepatol Bed Bench. 2012; 5: 79-83PubMed Google Scholar Therefore, we believe that the applied method is appropriate and could statistically exclude the confounding effects of residence, region, and unmeasured variables related to them. Consistent with our findings, Liu and colleagues3Liu S Zhang Y Ma R et al.Long-term exposure to ozone and cardiovascular mortality in a large Chinese cohort.Environ Int. 2022; 165107280Crossref Scopus (1) Google Scholar also observed an independent association between long-term exposure to ozone and cardiovascular mortality in China. Moreover, previous studies have suggested that the associations between ozone exposure and cardiovascular mortality is biologically plausible. For example, ozone inhalation might lead to a systemic inflammatory response, increased release of biomarkers related to coagulation, platelet dysfunction, and endothelial injury, and modulation of the autonomic nervous system and the neuroendocrine system. We have provided these discussions in our Article.1Niu Y Zhou Y Chen R et al.Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study.Lancet Planet Health. 2022; 6: e496-e503Summary Full Text Full Text PDF PubMed Scopus (5) Google Scholar We declare no competing interests. HK and MZ contributed equally to this Correspondence. Supplementary Material Download .pdf (.46 MB) Help with pdf files Supplementary appendix Download .pdf (.46 MB) Help with pdf files Supplementary appendix Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort studyThis study is, to the best of our knowledge, the first nationwide cohort study to show that long-term ozone exposure contributes to elevated risks of cardiovascular mortality, particularly from ischaemic heart disease, in a middle-income setting. The exposure–response function generated from this study could potentially inform future air quality standard revisions and environmental health impact assessments. Full-Text PDF Open Access