The higher temperature and ultraviolet, the lower COVID-19 prevalence-meta-regression of data from large US cities.

To the Editor:

Higher temperature and ultraviolet (UV) index in Northern Europe have been reported as the most important meteorological protective factors for the transmission of influenza virus. On the other hand, a recent study in China suggests that higher temperature and UV radiation may not be associated with a decrease in the epidemics of coronavirus disease 2019 (COVID-19). To determine whether prevalence of COVID-19 is modulated by meteorological conditions, we herein conducted meta-regression of data from large US cities.

We selected 33 large US cities with a population of >500,000 in 2010 from US Census Bureau (http://www.census.gov). We obtained (1) integrated number of confirmed COVID-19 cases in the county (to which the city belongs) on 14 May 2020 from Johns Hopkins Coronavirus Resource Center (https://coronavirus.jhu.edu), (2) estimated population in 2019 in the county from US Census Bureau, and (3) monthly meteorological conditions in the city for 4 months (from January to April 2020) from National Weather Service (https://www.weather.gov), World Weather Online (https://www.worldweatheronline.com), and Global Solar Atlas (https://globalsolaratlas.info/map) (Table 1 ). As the meteorological conditions, (1) mean temperature (F), total precipitation (inch), mean wind speed (mph), mean sky cover, and mean relative humidity (%) were available from National Weather Service; (2) mean pressure (mb), mean UV index, and total sun hours were obtainable from World Weather Online; and (3) total solar direct normal irradiation (DNI) (kWh/m2) in the average year was procurable from Global Solar Atlas. Monthly data for the 4 months (mean pressure/UV index and total sun hours were available for 3 months, from January to March 2020) were averaged or cumulated. The COVID-19 prevalence was defined as the integrated number of COVID-19 cases divided by the population. Random-effects meta-regression was performed by means of OpenMetaAnalyst (http://www.cebm.brown.edu/openmeta/index.html). In a meta-regression graph, the COVID-19 prevalence (plotted as the logarithm transformed prevalence on the y-axis) was depicted as a function of a given factor(plotted as meteorological condition on the x-axis).

Table 1

Extracted data in each large US city and county to which the city belongs

City	State	County	Covid-19 prevalence in the county			Meteorological parameter in the city
			Cases (n)	Population (n)	Prevalence	Mean temperature (F)	Total precipitation (inch)	Mean wind speed (mph)	Mean sky cover	Mean relative humidity (%)
Albuquerque	New Mexico	Bernalillo	1,149	679,121	0.00169	47.1	1.65	8.4	0.45	43
Austin	Texas	Travis	2,345	1,273,954	0.00184	60.6	12.06	8.2	0.63	72
Baltimore	Maryland	Baltimore	4,290	827,370	0.00519	46.5	14.66	7.4	0.67	61
Boston	Massachusetts	Suffolk	15,881	803,907	0.01975	40.6	12.62	12.1	0.62	58
Charlotte	North Carolina	Mecklenburg	2,342	1,110,356	0.00211	53.7	21.33	7.3	0.62	63
Chicago	Illinois	Cook	58,457	5,150,233	0.01135	37.9	10.86	10.3	–	–
Columbus	Ohio	Franklin	4,227	1,316,756	0.00321	41.6	19.31	8.9	0.75	67
Dallas	Texas	Dallas	6,837	2,635,516	0.00259	57.0	17.53	11.1	0.65	70
Denver	Colorado	Denver	4,359	727,211	0.00599	38.0	2.82	10.1	0.57	57
Detroit	Michigan	Wayne	18,770	1,749,343	0.01073	37.5	11.35	9.6	0.81	73
D.C.	D.C.	–	6,736	705,749	0.00954	48.7	14.61	9.1	0.70	63
El Paso	Texas	El Paso	1,456	839,238	0.00173	57.1	3.07	8.7	0.45	40
Fort Worth	Texas	Tarrant	4,076	2,102,515	0.00194	57.0	17.53	11.1	0.65	70
Houston	Texas	Harris	8,817	4,713,325	0.00187	64.2	13.90	8.5	0.68	71
Indianapolis	Indiana	Marion	7,793	964,582	0.00808	41.4	15.61	10.5	0.74	73
Jacksonville	Florida	Duval	1,215	957,755	0.00127	65.2	6.36	7.4	0.60	68
Las Vegas	Nevada	Clark	5,045	2,266,715	0.00223	58.1	2.31	6.5	0.45	37
Los Angeles	California	Los Angeles	35,392	10,039,107	0.00353	61.0	7.17	7.1	0.48	62
Louisville	Kentucky	Jefferson	1,741	766,757	0.00227	47.9	16.28	8.6	0.75	62
Memphis	Tennessee	Shelby	3,542	937,166	0.00378	52.9	27.76	8.8	0.68	71
Milwaukee	Wisconsin	Milwaukee	4,387	945,726	0.00464	35.4	10.52	10.4	0.68	66
Nashville	Tennessee	Davidson	3,745	694,144	0.00540	51.0	23.11	7.8	0.70	64
New York City	New York	New York City	188,545	8,336,817	0.02262	44.4	12.74	6.5	–	57
Oklahoma City	Oklahoma	Oklahoma	1,013	797,434	0.00127	49.3	10.86	12.1	–	–
Philadelphia	Pennsylvania	Philadelphia	19,093	1,584,064	0.01205	45.1	12.79	9.8	0.67	60
Phoenix	Arizona	Maricopa	6,599	4,485,414	0.00147	63.6	3.55	5.9	–	42
Portland	Oregon	Multnomah	940	812,855	0.00116	47.7	12.35	7.2	0.68	71
San Antonio	Texas	Bexar	1,976	2,003,554	0.00099	63.0	7.35	8.3	0.65	66
San Diego	California	San Diego	5,391	3,338,330	0.00161	61.0	6.69	5.1	0.55	69
San Francisco	California	San Francisco	1,999	881,549	0.00227	55.0	3.86	9.3	0.54	69
San José	California	Santa Clara	2,391	1,927,852	0.00124	55.2	4.06	5.8	0.51	65
Seattle	Washington	King	7,290	2,252,782	0.00324	46.4	18.15	9.0	0.73	73
Tucson	Arizona	Pima	1,696	1,047,279	0.00162	59.7	2.09	6.8	0.11	40

Results of the meta-regression were summarized in Table 2 . A slope of the meta-regression line was significantly negative for mean temperature (coefficient, −0.069; P < .001; Fig 1 , upper panel), mean UV index (coefficient, −0.445; P < .001; Fig 1, middle panel), total sun hours (coefficient, –0.002; P = .028; Fig 1, lower panel), and total solar DNI (coefficient, –0.002; P = .023; Fig 2 , upper panel), which indicated that COVID-19 prevalence decreased significantly as temperature, UV index, sun hours, and solar DNI increased. Whereas, the slope was significantly positive for mean wind speed (coefficient, 0.174; P = .027; Fig 2, middle panel) and sky cover (coefficient, 2.220; P = .023; Fig 2, lower panel), which indicated that COVID-19 prevalence increased significantly as wind speed and sky cover increased.

Table 2

Meta-regression summary

Covariate	Coefficient			P value
		Lower bound	Upper bound
Mean temperature (F)	−0.069	−0.093	−0.045	<.001
Total precipitation (inch)	0.038	−0.004	0.081	.075
Mean wind speed (mph)	0.174	0.020	0.328	.027
Mean sky cover	2.220	0.313	4.128	.023
Mean relative humidity (%)	0.007	−0.020	0.035	.613
Mean pressure (mb)	0.061	−0.220	0.342	.668
Mean ultraviolet index	−0.445	−0.585	−0.306	<.001
Total sun hours	−0.002	−0.004	−0.000	.028
Total solar direct normal irradiation (kWh/m2)	−0.002	−0.004	−0.000	.023

Fig 1

Meta-regression graph depicting the COVID-19 prevalence (plotted as the logarithm transformed prevalence on the y-axis) as a function of a given factor (plotted as a meteorological condition on the x-axis). Upper panel, mean temperature; middle panel, mean UV index; lower panel, total sun hours.

Fig 2

Meta-regression graph depicting the COVID-19 prevalence (plotted as the logarithm transformed prevalence on the y-axis) as a function of a given factor (plotted as a meteorological condition on the x-axis). Upper panel, total solar DNI; middle panel, mean wind speed; lower panel, sky cover.

The present meta-regression suggests that temperature, UV index, sun hours, and solar DNI may be negatively, and wind speed and sky cover may be positively associated with COVID-19 prevalence. Higher sun hours/solar DNI and lower sky cover are probably related to higher UV radiation. Despite the association of lower temperature and UV-index with the influenza transmission, no association of temperature and UV radiation with the COVID-19 epidemics has been reported, however, which may be denied by the present results of the association of higher temperature/UV index/sun hours/solar DNI and lower sky cover with lower COVID-19 prevalence. In conclusion, higher temperature/UV index/sun hours/solar DNI and lower wind speed/sky cover may be associated with lower COVID-19 prevalence (ie, lower temperature/UV index/sun hours/solar DNI and higher wind speed/sky cover may be associated with higher COVID-19 prevalence), which should be confirmed by further epidemiological researches adjusting for various risk and protective factors (in addition to meteorological conditions) of COVID-19.