Results
The first case was reported in NSW on 22 January 2020. For this case and
the subsequent 5 cases, infections were determined to have been acquired
overseas. The first case in which infection was determined to have been
acquired locally was notified on 26 February (NSW Government, 2020b).
Between 26 February and 31 March, a total of 749 cases were notified in
which infection was determined to have been acquired locally and which
reported a residence postcode. The median (interquartile range
[IQR]) day of reporting was 24 March (19 March, 27 March). Cases
were notified from 58 different local government areas, and 74% were
from the Greater Sydney area. The daily counts gradually increased until
reaching a peak around 27 March (Figure 1a) and then gradually
decreased. Climatic data was acquired from 52 weather observation
stations (Australian Government Bureau of Meteorology, 2020). During the
period 14 days prior to the first case with local transmission (12
February) to 31 March (49 days), 9am and 3pm relative humidity and 3pm
temperature remained approximately constant (median [IQR]: 82.0%
[75.5, 88], 60.0% [52.5, 69] and 24.1°C [21.9, 25.5],
respectively); however, the 9am temperature time-series (median
[IQR]: 19.9°C [18.4, 21.6]) showed a decreasing trend
(-0.106/day; r2 0.5109). During this period, there
were 27 days (55%) with recorded rainfall (median [IQR]: 1.5 mm
[0.2, 4.7]). Substantial rainfall (28−71 mm daily medians) was
reported between 7 and 10 February. The difference in 9am and 3pm
relative humidity difference varied over a wide range, whereas 9am and
3pm temperature difference remained relatively constant about a mean
value of 3.95°C (Figure 1b).
Collinearity was identified between some of the predictor variables, so
univariate GAMs were fit to cases and rainfall, 9am temperature, 9am
relative humidity, and the difference between 9am and 3pm relative
humidity. A negative relationship between cases and 9am relative
humidity was found, and a positive relationship between cases and 9am
temperature (Table 1). In multivariate analysis, only 9am relative
humidity was found to be significantly (P = 0.0304) associated
with cases (AIC 211.8; Table 2). The interaction between 9am temperature
and 9am relative humidity was not statistically significant. Using
either a 10-day or 21-day exponential moving average, the best-fitting
model (AIC 212.2 and 211.4, respectively) remained 9am relative humidity
only (P = 0.0408 and 0.0254, respectively). Including 9am
temperature in neither of these three models did not improve the model
fit.