Quantifying the effect of media limitations on outbreak data in a global online web-crawling epidemic intelligence system, 2008–2011

Figures & data

Fig. 1 Time series by disease and feed for top diseases by number of events in the HealthMap database. Note: influenza includes both untyped influenza and ‘swine flu’ H1N1 events.

Number of HealthMap Alerts by Disease, July 2008–July 2011.

Fig. 2 Breaking disease¹ alerts by week over the study period.

1. The following diseases are excluded from this series: influeza influenza/H1N1, cholera, avian influenza and foodborne illness.

Fig. 3 Natural log of the number of HealthMap alerts by day of the week.

Note: the rectangle marks 25th and 75th percentile distributions; the middle line is the median; whiskers mark the upper adjacent line (which represents 1.5 times the interquartile range); the dots represent extreme values and outliers.

Table 1. The influence of day of the week on log of total disease alerts*

	Google News			Moreover
Weekday (N=1,097)	% Change**	95% CI**	p	% Change**	95% CI**	p
Monday	−14.7	(−20.3, −9.1))	0.02	−9.2	(−14.6, −7.2)	0.04
Tuesday	Referent			Referent
Wednesday	1.4	(−3.1, 6.0)	0.84	−2.8	(−7.6, 1.1)	0.53
Thursday	7.8	(−3.7, 19.2)	0.26	1.5	(−6.7, 7.3)	0.75
Friday	−3.3	(−16.9, 10.3)	0.67	−3.2	(−11.9, 4.6)	0.48
Saturday	−58.3	(−72.7, −43.8)	<0.01	−39.8	(−48.9, −30.0)	<0.01
Sunday	−65.9	(−78.9, −52.9)	<0.01	−46.6	(−55.3, −34.8)	<0.01
Adjusted R^2			0.34			0.27

*Including influenza/H1N1, cholera, avian influenza and foodborne illness.

**Calculation based on formula from Halvorsen and Palmquist (38) applied to OLS coefficient; The Delta method (formula for which can be found in Greene (39))), was used in the computation of the confidence intervals.

OLS, ordinary least square.

Bold values indicate instances where p<0.05.

Table 2. The influence of holidays on log of total disease-related events by source (2008–2011)*

	Google News						Moreover
	Univariate			Multivariate			Univariate			Multivariate
International holidays (N=157)	% Change	95% CI	p	% Change	95% CI	p	% Change	95% CI	p	% Change	95% CI	p
International holidays (religious)
Christmas and New Years'	−27.4	(−41.3, −12.6)	<0.01	−27.4	(−41.4, −12.7)	<0.01	−19.7	(−38.7, −0.1)	0.08	−20.1	(−39.5, −0.6)	0.07
Easter	3.0	(−21.1, 26.5)	0.8				−7.7	(−42.3, 26.3)	0.66
Eid Al Adha	−15.6	(−49.6, 18.5)	0.4				−23.7	(−53.0, 5.8)	0.17
Eid Al Fitr	16.2	(−3.0, 36.3)	0.1				37.7	(8.7, 68.0)	<0.01	17.7	(−11.0, 46.5)	0.19
European holidays
Labor day (Europe)	−13.1	(−31.2, 5.5)	0.2				15.0	(−5.6, 35.8)	0.13
Europe day (May 9)	−9.5	(−33.0, 14.7)	0.5				−39.3	(−84.5, 5.8)	0.19
Armistice day	5.1	(−6.6, 16.9)	0.4				−52.8	(−113.4, 8.52)	0.26
US holidays
Martin Luther King day	37.7	(−34.8, 109.5)	0.2				12.7	(−10.2, 36.2)	0.25
Presidents day	13.9	(−8.4, 35.2)	0.2				19.7	(−15.4, 53.8)	0.24
Memorial day (US)	16.2	(−43.2, 74.9)	0.6				−36.2	(−62.2, −10.6)	0.03	−37.0	(−62.9, −10.9)	0.03
Independence day (July 4)	10.5	(−32.2, 52.4)	0.6				−17.3	(−48.8, 13.5)	0.32
Labor day (US)	12.7	(−27.9, 53.2)	0.5				64.9	(−19.3, 148.1)	0.06	51.7	(−36.0, 139.3)	0.16
Columbus day	−13.9	(−47.6, 20.3)	0.5				20.9	(−15.1, 56.8)	0.22
Veteran's day (US)	5.1	(−6.56, 16.9)	0.4				−52.8	(−113.3, 8.53)	0.26
Thanksgiving (US)	−33.6	(−47.0, −20.7)	<0.01	−34.9	(−47.7, −21.7)	<0.01	−41.1	(−67.9, −14.3)	0.02	−41.6	(−68.5, −14.5)	0.02
Other events:
US election (2008)	8.3	(−5.4, 22.8)	0.2				13.9	(−8.70, 37.29)	0.2
Adjusted R^2				0.05						0.05

*Calculation of percent change based on formula from Halvorsen and Palmquist (38) applied to OLS coefficients; Delta Method, as described in Greene (39), was used in the computation of 95% confidene interval.

OLS, ordinary least square.

Bold values indicate instances where p<0.05.

Table 3. Measurement of crowd out effect based on OLS specification (with Newey–West standard errors) for Google News and Moreover

Crowd out (N=155) Outcome = total disease stories	Covariate	Time	Coefficient	95% CI	p	Adj-R^2	AIC	Ljung-Box p
Google	Influenza	Lag(1)	−0.08	(−0.11, −0.03)	0.00	0.45	−0.60	0.18
	Cholera	Contemp.	−0.03	(−0.12, 0 .06)	0.51	0.42	5.23	0.22
	Avian influenza	Lead(1)	0.07	(−0.02, 0.15)	0.09	0.40	12.20	0.15
	Foodborne illness	Contemp.	0.02	(−0.03,0.07)	0.61	0.39	15.70	0.06
Moreover	Influenza	Lag(1)	−0.2	(−0.38, −0.01)	0.04	0.47	124.30	0.02
		Lag(2)	0.15	(−0.04, 0.32)	0.12
	Cholera	Lag(1)	0.02	(−0.06, 0.11)	0.67	0.46	124.30	0.27
	Avian influenza	Lag(2)	0.02	(−0.04,0.09)	0.55	0.45	124.20	0.26
	Foodborne illness	Contemp.	0.04	(−0.01,0.1)	0.48	0.46	125.00	0.21

OLS, ordinary least square.

Bold values indicate instances where p<0.05.

Halvorsen R, Palmquist R. The interpretation of dummy variables in semilogarithmic equations. Am Econ Rev. 1980; 70: 474–5.

 Web of Science ®Google Scholar

Greene W. Econometric analysis. 2011; 7th ed, Upper Saddle River (NJ): Prentice Hall.

 Google Scholar