Supplementary Materials to Pace-Makers Lost: Shifting Spatial Epidemiology of Reemerging Pertussis in Continental USA Marc Choisy MIVEGEC (UM1-UM2-CNRS 5290-IRD 224), Centre IRD, 911 avenue Agropolis BP 64501, 34394 Montpellier C´edex 5, France [email protected] Pejman Rohani Department of Ecology and Evolutionary Biology and Center for the Study of Complex Systems, University of Michigan, Ann Arbor, MI, USA Fogarty International Center, National Institutes of Health, Bethesda, MD 20892, USA [email protected]

1

1

Selection of the two analyzed eras

Given a long (1970–1990) and almost uninformative period of time characaterizing the US pertussis notification case time series, we performed our analyses on two separate eras, before and after this central part. The main concerns about selecting a portion of the data to perform our analysis are, first, the arbitrary choice and, second, the robustness of the results respective to this choice. We addressed these concerns by examining the behaviors of the distribution of the pairwise correlation coefficients (fig. S1A and B), the number of states above and below the CCS (fig. S1C and D), the spatial correlation functions (not shown because involving too many figures) and the global wavelet spectra (not shown because involving too many figures) calculated on the 1951-x (in red) and x-2010 (in blue) time periods when x varies between 1951 and 2010. All these analyses identified sharp transitions around 1963 and 2002, and we based the selection of the two analyzed eras on these transitions. Moreover, the results presented in the article for the first era with x =1962 were fairly robust with respect to x. Finally, selecting the time period before or after the wavelet decomposition did not affect significantly the results.

2

Decomposition of the signal

Fig. S2, page 4, illustrates the different steps of data processing from raw data to residual phase angles for the first (1951-1962, fig. S2A-G) and second eras (2002-2010, fig. S2H-N), as detailed in the Materials and Methods section of the main text. The raw data (fig. S2A and H) are first square-rooted in order to stabilize the variance compared to the mean (fig. S2B and I). Time series are then normalized (i.e. centered and reduced) in order to allow comparisons of qualitative features (i.e. periodicity and phase) between different states (fig. S2C and J). Times series are then filtered around the dominant period: between 3.5 and 4.5 years for the first era, fig. S2D and between 5 and 6 years for the the second era, fig. S2K. The phase of the filtered signal is calculated (fig. S2E and L) and “linearized” (fig. S2F and M). The aim of the linearization step is to transform phase from a circular function oscillating between −π and +π (fig. S2E and L) to a function increasing linearly from −π to +∞ (fig. S2F and M). This allows to calculate the “residual phase angles” which are simply the residuals of a linear model expressing the phases of all the states as a function of time (fig. S2G and N). This is an alternative of phase difference to express the timing of epidemics and its advantage compared to phase difference is that we don’t have to arbitrarily choose a reference time series to express these timings.

3

Visualisation of traveling wave on raw data

Phase calculation on filtered signal allows to reveal a conspicuous hierarchy in the timing of epidemics from the coasts to the inland (figure 3 of the main text). In order to visualise such a pattern on raw data, Fig. S3 plots the time series of pertussis incidence of each state (one state per line), where the states are ordered by longitude. To ease the qualitative comparison from state to state, the incidences of each state are transformed to lies between 2

B

0 5

correlation coefficient 0.0 0.2 0.4 0.6

35

1951−1980 1981−2010

number 15 25

A

proportion of months with no notification 0.0 0.2 0.4 0.6 0.8

C

● ●

D 1951−1980 1981−2010

● ● ● ● ●● ●●●● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ●●● ●●● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ●● ●● ● ●● ● ● ● ● ● ●● ●

0

1960

●

10000 20000 30000 population size x1000

number of states above CCS 0 10 20 30 40

0.0 0.2 0.4 0.6 0.8 correlation coefficient

before after 1980 2000 separating year

●●●●●●● ● ● ●●●●

before after

●● ● ● ● ● ●

●●● ●●●● ●●● ●●●● ●●● ●●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●● ●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

1960

1980 2000 separating year

Figure S 1: Variation of the pairwise correlation distribution and CCS over the 1951-x and x-2010 time periods when x varies between 1951 and 2010. (A) Distributions of the pairwise correlation coefficient for the particular case of x =1980. (B) Changes in the distributions when x varies between 1951 and 2010. The lines are the mean values of the distribution and the colored areas are the 50% confidence intervals. The vertical lines show the first days of 1963 and 2002. (C) CCS for the particular case of x =1980. (D) Changes in the number of states above the CCS when x varies between 1951 and 2010. The vertical lines show the first days of 1963 and 2002.

3

F

G

0

5 10 20 phase angle

−3

E −1 1 2 3 −0.5 0.5 1.0 −2 0 2 4 phase angle filtered component norm. incid.

phase angle filtered component norm. incid. −1 1 2 3−1.5 −0.5 0.5 1.5−2 0 2 4 6

D

−3 −1 1 2 3 residual phase angle

−3

C 60

0

10 20 30 incidence

0

incidence 10 20 30 40 0

B

residual phase angle phase angle −3 −1 1 2 3 0 5 10 20

400 800 incidence

incidence 500 1500

A H

I

1952 1954 1956 1958 1960 1962 2002 2004 2006 2008 2010 year year

Figure S 2: Decomposition of the signal. See text for explanations.

4

J

K

L

M

N

W 1952

1956

1960 year

2003

2007

W 2011

0.0

10 20 30 40 states ordered by longitude

E

states ordered by longitude 10 20 30 40

E

0.2 0.4 0.6 0.8 1.0 relative number of cases

0 and 1 (and called “relative number of cases”). One can thus guess in the first era of fig. S3 the traveling wave that is much more clearly visible and quantified on figure 3 of the main text. There is no such clear structure in the second era.

Figure S 3: Relative numbers of notification cases for each of the 49 states ordered according to the longitude of their population centers from west (bottom) to east (top) over the periods 1951–1962 (left panel) and 2002– 2010 (right panel). States are: OR (line 1), WA (2), CA (3), NV (4), ID (5), UT (6), AZ (7), MT (8), WY (9), NM (10), CO (11), ND (12), SD (13), NE (14), TX (15), OK (16), KS (17), MN (18), IA (19), AR (20), MO (21), LA (22), MS (23), WI (24), IL (25), AL (26), TN (27), IN (28), KY (29), MI (30), GA (31), OH (32), FL (33), SC (34), WV (35), NC (36), VA (37), PA (38), DC (39), MD (40), DE (41), NY (42), NJ (43), CT (44), VT (45), NH (46), RI (47), MA (48), ME (49).

4

Movies of the pertussis spatial dynamics

Two movies of the spatial dynamics of pertussis in the US can be seen at http://marcchoisy.free.fr/pertussis. In each movie, the upper panel shows the time series of the pertussis incidence (number of reported new cases divided by population size) aggregated for the all USA. The vertical blue bar refers to the time point to which the map in the lower panel corresponds. The map of the first movie shows the values of the filtered (between periods of 3.5 and 4.5 years) time series for each state whereas the map of the second movie depicts spatially smoothed values, using a loess regression with the longitude, latitude (and interaction) of the centroids of each state as explanatory variables. These centroids are represented on the map by the black dots. These movies clearly illustrates the wave of pertussis propagation from the coasts towards inland.

5