Air pollution, SARS-CoV-2 incidence and COVID-19 mortality in Rome - a longitudinal study.

Chronic exposure to ambient air pollution has been related to increased mortality in the general population [1]. After the outbreak of SARS-CoV-2 pandemic in 2019, there has been a fast proliferation of epidemiological studies linking ambient air pollution to COVID-19 incidence or adverse prognosis [2]. It has been hypothesised that ambient air pollution might increase human vulnerability to viruses by reducing immune defences, promoting a low-level chronic inflammatory state, or leading to chronic diseases [3]. Most studies applied ecological designs, and failed to account for key individual-level or area-level determinants of COVID-19 spread or severity, such as demographic characteristics of the studied populations, socioeconomic or clinical susceptibility, and area-level proxies of disease spread such as mobility or population density [4].

In this study we aimed to investigate the association between chronic exposure to air pollution and SARS-CoV-2 incidence and COVID-19 mortality, independent from age, sex, individual-level and area-level socio-economic deprivation, clinical history, and neighborhood characteristics.

All subjects 30+ year old resident in Rome, Italy, at January 1st 2020 were followed-up until April 15th 2021 through record linkage between the different administrative archives of the Lazio Region Health Information System: population and mortality registries, 2011 Census data, and COVID-19 surveillance system. The COVID-19 integrated surveillance system collects all the new confirmed SARS-CoV-2 infections reported to the Regional Service for Surveillance of Infectious Diseases (SERESMI) throughout the Lazio Region. Each subject entered at baseline and was observed until death, emigration out of the study area or end of follow-up, whichever came first. For each subject, baseline information was available on demographic characteristics (age, sex, marital status, place of birth), socio-economic indicators (education level, occupational status, census block-level deprivation), clinical history (a list of 67 prevalent conditions based on past 5-year hospitalisations and drug prescriptions), neighborhood characteristics (housing prices, unemployment rate, education level), and geographical coordinates of the residential address. Three study outcomes were defined: incidence of SARS-CoV-2 infection (newly identified cases based on a positive test through reverse transcription polymerase chain reaction), COVID-19 mortality (deaths within 30 days from infection), and non-COVID-19 mortality (deaths among non-cases, or after 30 days since infection). Annual average concentrations of particulate matter smaller than 2.5 micron (PM2.5) and nitrogen dioxide (NO2) were estimated for 2019 at 1-km2 spatial resolution for the entire Italian territory, using a machine learning spatiotemporal model which incorporated data from existing air quality monitoring networks, satellite images, atmospheric models, land-use and population characteristics [5].

We applied Cox proportional hazard models with adjustment for individual- and area-level

covariates. These include: calendar time (as time axis); demographic variables (age in five classes: 30–54, 55–64, 65–74, 75–84, 85+ years, sex, marital status, place of birth, nationality); socio-economic indicators (education level, occupational status and census block-level deprivation index); pre-existing chronic conditions (number of any conditions out of a list of 67 diseases, and six specific diseases); neighborhood-level characteristics. The latter were adjusted differently for incidence (a strata term for the 155 districts in Rome) or mortality outcomes (housing prices, unemployment rate and % university degree at the district level). We operated this choice because we assumed that factors related to viral spread in the general population (person-to-person contacts, individual mobility, etc.) were better accounted for by assuming different baseline rates for each neighborhood. Instead, socio-economic characteristics of the residential neighborhood could adequately adjust for spatially heterogenous susceptibility of the study population. Next, we added each exposure in turn (PM2.5 or NO2) as a linear term, and expressed all estimates as hazard ratios (HR), and 95% confidence intervals (95% CI), of the study outcomes, per increments in the exposures equal to their interquartile ranges (IQR).

We conducted a number of additional/sensitivity analyses: we defined three pandemic waves consistent with the viral spread in Rome (February-September 2020, October-December 2020, January-April 2021) and estimated wave-specific effects of air pollutants using time-varying models; we replaced the strata term for districts with population density in the adjustment model for SARS-CoV-2 incidence; we dropped the pre-existing conditions from the adjusted model, under the assumption that these might act as mediators, rather than confounders, of the studied associations; we adjusted for district-level rates of diabetes, COPD and lung cancer as proxies for BMI and smoking; we analyzed COVID-19 hospitalisations and intensive care units as alternative outcomes of COVID-19 severity; finally, we estimated the exposure-response functions between the air pollutants and the study outcomes by modelling air pollutants with natural splines.

Descriptive statistics and results are displayed in table 1. We enrolled 1,594,308 subjects, with a mean follow up of 461 days (sd=48 days). Of these, 79 976 individuals were infected with SARS-CoV-2, 2656 died within 30 days from infection, and 1002 died after 30 days from infection. 31 563 individuals died without ever being diagnosed with COVID-19 during the follow-up. The average air pollution exposures at the baseline were: 14.63 µg·m−3 for PM2.5 (IQR 0.92 µg·m−3) and 31.45 µg·m−3 for NO2 (IQR 9.22 µg·m−3). Infection incidence rates, and HRs form the fully adjusted model, were highest among younger subjects, students or employed people, those with higher socio-economic deprivation and in neighbourhoods with lowest housing prices. COVID-19 mortality rates and HRs substantially increased with age and number of pre-existing chronic conditions, and were higher among males, subjects with poor education and highest deprivation, or among patients with pre-existing renal failure, heart failure, ischemic heart disease, COPD, type 2 diabetes or cancer. Similar patterns emerged for non-COVID-19 mortality, although with reduced differentials by sex and socio-economic deprivation.

TABLE 1

Population characteristics, study outcomes, and associations between individual covariates, exposure and the study outcomes in the fully adjusted model

		Study population (n=1,594,308)		Incident cases (n=79 976)				COVID19 deaths  (n=2656)				non-COVID19 deaths (n=32 565)
		N	%	rate* 1000 p-y	HR	95% CI		rate* 1000 p-y	HR	95% CI		rate* 1000 p-y	HR	95% CI
Demographic variables
Age (years)	30–54	699,791	43.9	47.1	1.00	–	–	0.09	1.00	–	–	1.19	1.00	–	–
	55–64	332,651	20.9	41.1	0.84	0.82	0.85	0.46	3.76	2.89	4.90	4.57	2.95	2.74	3.19
	65–74	260,430	16.3	30.2	0.62	0.60	0.64	1.55	8.93	6.93	11.53	13.42	5.74	5.33	6.18
	75–84	208,644	13.1	28.3	0.59	0.57	0.61	3.79	16.85	12.92	21.99	38.65	11.81	10.94	12.75
	85+	92,792	5.8	31.3	0.67	0.64	0.71	8.05	34.94	26.61	45.90	136.02	37.51	34.71	40.54
Sex	Male	715,610	44.9	42.3	1.00	–	–	1.76	1.00	–	–	16.56	1.00	–	–
	Female	878,698	55.1	37.8	0.94	0.92	0.95	0.96	0.46	0.43	0.51	15.55	0.81	0.79	0.83
Marital status	Single	724,491	45.4	39.6	1.00	–	–	1.32	1.00	–	–	16.15	1.00	–	–
	Married	682,340	42.8	40.0	1.01	0.99	1.02	1.32	1.02	0.94	1.11	15.63	1.00	0.97	1.02
	Separated/divorced	67 373	4.2	39.3	1.00	0.96	1.03	1.30	0.95	0.79	1.16	17.22	1.05	1.00	1.11
	Widowed	120,104	7.5	39.8	1.00	0.98	1.03	1.29	0.96	0.82	1.12	16.58	1.02	0.97	1.06
Born in Rome	Yes	940,690	59.0	39.7	1.00	–	–	1.29	1.00	–	–	15.67	1.00	–	–
	No	653,618	41.0	40.0	0.99	0.97	1.00	1.35	0.99	0.91	1.08	16.49	1.00	0.98	1.03
Italian nationality	Yes	1,380,875	86.6	39.8	1.00	–	–	1.31	1.00	–	–	15.83	1.00	–	–
	No	213,433	13.4	39.8	1.00	0.98	1.02	1.37	1.04	0.92	1.18	17.18	1.01	0.98	1.05
Socio-economic variables
Level of education	Primary or less	195,587	12.3	35.5	1.00	–	–	4.40	1.00	–	–	54.72	1.00	–	–
	Middle school	350,961	22.0	42.1	0.95	0.93	0.98	1.49	0.85	0.77	0.94	17.12	0.93	0.91	0.96
	High school	652,376	40.9	41.9	0.93	0.91	0.96	0.77	0.77	0.69	0.86	9.13	0.84	0.82	0.87
	University or more	395,384	24.8	36.4	0.86	0.83	0.88	0.60	0.65	0.56	0.75	7.78	0.77	0.74	0.80
Employment status	Employed	891,706	55.9	44.7	1.00	–	–	0.40	1.00	–	–	3.62	1.00	–	–
	Searching first empl.	18 999	1.2	39.1	0.80	0.75	0.85	0.17	0.80	0.30	2.15	2.53	1.27	0.99	1.64
	Unemployed	59 121	3.7	39.3	0.81	0.78	0.84	0.36	1.17	0.79	1.73	3.73	1.39	1.23	1.57
	Retired	317,698	19.9	28.9	0.82	0.80	0.85	4.24	1.21	1.05	1.39	53.43	1.43	1.36	1.49
	Student	40 097	2.5	42.7	0.94	0.90	0.98	0.10	0.86	0.35	2.09	1.10	0.77	0.59	1.00
	Housewife	181,316	11.4	35.6	0.87	0.84	0.89	1.24	1.09	0.92	1.30	18.05	1.21	1.15	1.28
	Other	85 371	5.4	35.7	0.82	0.79	0.84	1.95	1.42	1.18	1.69	24.85	1.71	1.62	1.80
Socioeconomic deprivation (of the census block)	Low	338,918	21.3	34.4	1.00	–	–	1.10	1.00	–	–	16.39	1.00	–	–
	Mid-low	424,703	26.6	37.0	1.01	0.99	1.03	1.30	1.16	1.03	1.30	15.92	1.02	0.99	1.05
	Medium	290,133	18.2	40.7	1.01	0.99	1.04	1.23	1.11	0.97	1.27	15.55	1.06	1.03	1.10
	Mid-high	251,250	15.8	44.7	1.05	1.02	1.08	1.38	1.22	1.06	1.41	14.87	1.06	1.02	1.10
	High	289,304	18.1	45.2	1.07	1.04	1.09	1.63	1.31	1.14	1.50	17.14	1.13	1.09	1.17
Pre-existing chronic conditions #
Number	0	763,781	47.9	41.6	1.00	–	–	0.24	1.00	–	–	2.80	1.00	–	–
	1	348,492	21.9	39.0	1.07	1.05	1.09	1.00	1.62	1.37	1.91	11.26	1.44	1.37	1.51
	2	209,386	13.1	37.0	1.12	1.09	1.15	1.70	1.74	1.47	2.06	22.37	1.65	1.57	1.73
	3	121,952	7.6	35.8	1.13	1.10	1.17	2.82	2.15	1.80	2.56	34.47	1.84	1.74	1.93
	4+	150,697	9.5	39.3	1.24	1.19	1.29	5.98	2.79	2.33	3.33	73.08	2.19	2.08	2.31
Specific conditions	Cancer	41 985	2.6	37.9	1.06	1.01	1.11	4.13	1.47	1.27	1.70	97.84	3.47	3.37	3.58
	Type 2 diabetes	101,874	6.4	40.3	1.08	1.05	1.12	4.16	1.13	1.02	1.25	43.72	1.06	1.03	1.09
	Ischemic heart disease	65,307	4.1	40.9	1.08	1.03	1.12	6.31	1.10	0.98	1.23	70.78	1.06	1.02	1.09
	Heart failure	60,726	3.8	40.5	1.11	1.06	1.16	7.58	1.33	1.19	1.48	104.39	1.52	1.48	1.57
	COPD	88,007	5.5	38.7	1.07	1.03	1.10	4.57	1.15	1.04	1.28	62.88	1.33	1.30	1.37
	Renal failure	21,850	1.4	42.2	1.14	1.07	1.21	10.27	1.65	1.44	1.89	128.68	1.55	1.49	1.61
District-level covariates
House prices (quintiles)	1	338,871	21.3	49.5	–	–	–	1.23	0.85	0.66	1.10	12.14	0.94	0.87	1.01
	2	330,823	20.8	43.3	–	–	–	1.45	0.79	0.63	0.98	16.24	0.94	0.89	1.00
	3	300,216	18.8	38.3	–	–	–	1.47	0.86	0.70	1.05	17.65	0.98	0.93	1.04
	4	314,726	19.7	34.8	–	–	–	1.24	0.84	0.72	0.98	17.18	0.99	0.94	1.03
	5	309,672	19.4	31.9	–	–	–	1.20	1.00	–	–	17.21	1.00	–	–
Unemployment rate¶	mean, IQR	6.5	1.8	–	–	–	–	–	1.04	0.93	1.16	–	1.02	0.99	1.05
% university degree or more¶	mean, IQR	39.3	31.3	–	–	–	–	–	0.86	0.71	1.04	–	1.05	1.00	1.11
Exposures ¶
PM2.5 (μg·m−3)	mean, IQR	14.63	0.92	–	1.01	0.99	1.03	–	1.08	1.03	1.13	–	1.01	1.00	1.02
NO2 (μg·m−3)	mean, IQR	31.45	9.22	–	1.00	0.98	1.02	–	1.09	1.02	1.16	–	1.02	1.00	1.04

Rates are computed as ratios between numbers of outcomes and person-years, multiplied by 1000. Hazard Ratios are estimated from a Cox proportional hazards model adjusted for calendar time (time axis), age (five classes), sex, marital status (for classes), place of birth, Italian nationality, education level (four classes), employment status (seven classes), census-block-level socioeconomic deprivation index (five classes), number of pre-existing conditions (five classes), presence of six specific pre-existing conditions (cancer, type-2 diabetes, ischemic heart disease, heart failure, COPD, renal failure), and district-level characteristics. The latter are adjusted with a “strata” term for the 155 Rome districts in incidence analysis, and with three district-level covariates (house prices in five classes, % university degree or more, unemployment rate) in the mortality analyses. PM2.5: particulate matter smaller than 2.5 micron; NO2: nitrogen dioxide; IQR: interquartile range; HR: hazard ratio; CI: confidence interval.

#Pre-existing chronic conditions include a list of 67 groups of diseases based on past 5-year hospital admissions or drug prescriptions.

¶Associations between continuous covariates (unemployment rate, % university degree or more) and air pollutants with the study outcomes are expressed as HRs (and 95% CI) per IQR increments. Exposures are modelled one at a time (single-pollutant models).

Table 1 reports the results of the association between PM2.5 and NO2 with the three study outcomes in the fully adjusted model. We found no association between air pollution and SARS-CoV-2 incidence: IQR increments in PM2.5 and NO2 were associated with HRs of 1.01 (0.99, 1.03) and 1.00 (0.98, 1.02), respectively. In contrast, we estimated strong associations between the two air pollutants and COVID-19 mortality: IQR increments in PM2.5 and NO2 were associated with HRs of 1.08 (1.03, 1.13) and 1.09 (1.02, 1.16). The association between air pollutants and non-COVID-19 mortality was comparatively smaller than the one with COVID-19 mortality: we estimated HRs of 1.01 (1.00, 1.02) and 1.02 (1.00, 1.04) per IQR increments in PM2.5 and NO2, respectively. The results of the additional/sensitivity analyses confirm the main findings: associations were unaffected by alternative adjustment models, they did not differ substantially by pandemic wave, and were significant with hospital admissions but not with accesses to intensive care units. Finally, the exposure-response functions were consistent with flat associations between the two air pollutants and SARS-CoV-2 incidence, while associations with mortality outcomes were approximately linear, and much steeper for COVID-19 mortality (data not shown).To date, few studies investigated the relationship between air pollution and COVID-19-related outcomes in population-based longitudinal studies. Chadeau-Hyam et al. (2020) and Elliott et al. (2021) linked COVID-19 data and mortality records to the UK Biobank and found no association between residential PM2.5 exposure and either positive testing to SARS-CoV-2 [6] or COVID-19 mortality [7], after multivariate adjustment for individual and area-level risk factors. Similarly, no association between air pollutants and SARS-CoV-2 positive testing was detected in the COVICAT cohort of Catalonia, Spain, although a statistically significant association was estimated with severe COVID-19 disease among infected patients [8]. No association between PM2.5 or NO2 and mortality was found in a prospective longitudinal study conducted in Ontario, Canada, while significant associations were estimated with hospitalisations and accesses to intensive care units [9]. In contrast, positive associations between PM2.5 exposure and COVID-19 incidence were estimated in northern Italy [10] and southern California [11].

Our estimates of association between air pollutants and COVID-19 mortality are similar to those found in previous large ecological studies [2, 12], and much higher than those usually found with natural-cause mortality in the general population [1, 13]. Several mechanisms have been proposed as responsible for an enhanced severity of COVID-19 in combination with exposure to air pollution. First, air pollution-induced inflammation may amplify inflammation due to COVID-19 and lead to adverse health outcomes, including premature death; second, air pollution may reduce the immune response against the virus by inhibiting phagocytic function of macrophages and decreasing the T-cell response; third, chronic exposure to air pollution may induce endothelial damage and microthrombi, thus increasing the risk of cerebral damage, pulmonary embolism, and cardiac dysfunction among COVID-19 patients [14].

This study has two main limitations. First, our cohort lacks data on relevant individual-level lifestyle characteristics such as smoking, physical activity and dietary habits, or physiological parameters such as body-mass index and cholesterol levels. While these might confer greater susceptibility to the individuals, it is however not clear to what extent they should correlate with ambient air pollution, once area-specific covariates (e.g., socio-economic deprivation) are accounted for. Secondly, our COVID-19 surveillance system, especially in the early stages of the pandemic, could only identify a selected sample of all infected individuals, e.g., those with severe symptoms or close contacts of primary cases. The testing policy was broadened to asymptomatic primary contacts and to various screening programmes (e.g., ahead of hospital admission for other causes) only after the first wave, when Italy entered the transition phase and a test–track–trace strategy was adopted. Therefore, our definition of SARS-CoV-2 incidence is only partial. Again, however, there are no a priori reasons to believe that included and excluded cases should be different with regard to air pollutant distributions.

In conclusion, in this large longitudinal study, long-term residential exposure to air pollution was associated with increased mortality among COVID-19 patients, but not with SARS-CoV-2 incidence in the general population. Our study supports the hypothesis that chronic exposure to air pollution might increase human vulnerability to viruses, thus worsening prognosis of COVID-19 cases while they are unlikely to increase the spread of infection in the general population.