Rainfall drives hydrocephalus in East Africa

Clinical article

Restricted access

Object

Hydrocephalus is one of the most common brain disorders in children throughout the world. The majority of infant hydrocephalus cases in East Africa appear to be postinfectious, related to preceding neonatal infections, and are thus preventable if the microbial origins and routes of infection can be characterized. In prior microbiological work, the authors noted evidence of seasonality in postinfectious hydrocephalus (PIH) cases.

Methods

The geographical address of 696 consecutive children with PIH who were treated over 6 years was fused with satellite rainfall data for the same time period. A comprehensive time series and spatiotemporal analysis of cases and rainfall was performed.

Results

Four infection-onset peaks were found to straddle the twice-yearly rainy season peaks, demonstrating that the infections occurred at intermediate levels of rainfall.

Conclusions

The findings in this study reveal a previously unknown link between climate and a neurosurgical condition. Satellite-derived rainfall dynamics are an important factor in driving the infections that lead to PIH. Given prior microbial analysis, these findings point to the importance of environmental factors with respect to preventing the newborn infections that lead to PIH.

Abbreviations used in this paper:NOAA = National Oceanic and Atmospheric Administration; PIH = postinfectious hydrocephalus.

Article Information

Address correspondence to: Steven J. Schiff, M.D., Center for Neural Engineering, W311 Millennium Science Complex, Penn State University, University Park, Pennsylvania 16802. email: sschiff@psu.edu.

Please include this information when citing this paper: published online July 6, 2012; DOI: 10.3171/2012.5.PEDS11557.

© AANS, except where prohibited by US copyright law.

Headings

Figures

  • View in gallery

    Panel Aa shows a satellite 0.1° × 0.1° grid overlain on the geographic districts of Uganda. Monthly rainfall was averaged within each district, and patients with PIH were assigned to a district. Satellite data are missing for January and February of 2000, and these data were set to 0. These 77 districts were mapped to a 10 × 10 adjacency matrix in panel Ab, keeping at least 1 or more neighboring districts in contact to facilitate display of spatiotemporal analysis. The 23 black squares are empty. The rainfall by month from each district is shown overlain in panel Ba, and the spectrogram demonstrates a strong twice-yearly frequency of rainy seasons. Birth, febrile illness, and surgery dates are shown in panels Bb, Bc, and Bd, respectively, and their spectrograms are complex. Note that the time axes for the spectrograms are shortened to 1.5–4.5 years, reflecting the resolution within the 3-year sliding windows used. Summing the total rainfall (in mm) for each district over all 6 years reflects the high levels of rainfall across Uganda in panel Ca and the more complex spatial distribution of cases over all 6 years in panel Cb (a base 10 log scale is used for cases to facilitate the display). A linear regression of total rainfall to cases (using febrile illness dates, for comparison later) in panel Cc demonstrates a modest fit (coefficient of determination, R2 = 0.44), and the residuals of this fit steadily increase as total rainfall increases, indicating that the relationship between rain and cases over geography is not a simple one explained by total rainfall. A regression by birth dates is nearly identical (R2 = 0.43).

  • View in gallery

    Average global time series (univariate) are shown in panels A and B, and spatiotemporal analysis in panels C–E. After averaging rainfall and summing cases across districts by month, their cross-correlations are shown in panel A. Confidence bounds were calculated as described in Methods, representing twice the SD of surrogate correlations displayed as the dotted line bootstrap confidence limits in panel A. All correlations greater than these confidence limits are potentially significant, especially where consecutive significant excursions are seen. The correlation with rainfall is greatest for febrile illness dates, and the surgery dates are offset from febrile illness and birth dates by several months, as expected. The coherograms of the data from panel A are shown in B, where a prominent band at 4 cycles per year emerges from each of the data sets. The full data sets for each district were decomposed using a singular value decomposition, and the eigenvalues (statistical importance of the eigenimages) and eigenimages (most statistically common pattern to the data, ordered by the eigenvalues from largest to smallest) are shown for rainfall and febrile illness dates in panel C. There is a prominent first mode, which represents the average total rainfall or case numbers, followed by second and third modes displaying fluctuations about the mean. These modes were used as filters to form reconstructed univariate time series in panel D (products of eigenvalue, modal amplitude, and eigenimage, and summing to reconstruct data), where it is seen that there were no correlations between the first rainfall mode and case data. Testing all combinations of the first 6 modes, we find the largest correlations by summing modes 2 and 3, where febrile illness dates demonstrate larger correlations with rainfall than birth dates, and no correlation in these modes is seen for surgery dates. In panel E we show coherograms from the sum of modes 2 and 3, demonstrating that there is very substantial correlation between birth and febrile illness dates and that febrile illness dates correlate more strongly with surgery dates (in the correlation sums at arbitrary lags [Csum]) than do the birth dates. C0 = absolute value of the total correlation at 0 time lag when greater than the 2-SD confidence limit; Csum = sum of the total significant correlation at all lags.

  • View in gallery

    Histograms for all district case data per month (5544 samples) shown as a function of rainfall per month for febrile illness, birth dates, and surgery dates in panels Aa–Ac. The cases are strongly peaked at intermediate levels of rainfall. Normal distribution fits are shown, which indicate that the birth and febrile illness months have almost identical means, whereas the surgery mean rainfall values are offset. A histogram of monthly rainfall is shown in panel Ba, along with monthly cases based on febrile illness, birth, and surgery dates in panels Bb–Bd. Tests of nonuniformity based on Shannon entropy show that each case histogram is significantly nonuniform with respect to monthly rainfall, with febrile dates being the most nonuniform (p = 0.0066). Note the 4 peaks per year in monthly febrile cases flanking the 2 peaks in rainfall in April and October. Rainfall for each district was then narrow-band filtered from 1.5 to 2.5 Hz by using a fifth-order Chebyshev filter applied so as not to distort phase in panel Ca, and a Hilbert transform was used to assign phase from 0 to 4π to represent yearly cycles in panel Cb. The average rainfall per phase was fit by a sine wave in panel Da, and the histograms of case data are shown in panels Db–Dd. Again, assigning phase values rather than calendar months demonstrates that febrile illness occurs with 4 peaks per year, straddling the phase peaks, which coincide with peak rainfall months.

References

1

Aziz IA: Hydrocephalus in the Sudan. J R Coll Surg Edinb 21:2222241976

2

Bendat JSPiersol AG: Random Data: Analysis and Measurement Procedures ed 2New YorkWiley1986. 484516

3

Chi JHFullerton HJGupta N: Time trends and demographics of deaths from congenital hydrocephalus in children in the United States: National Center for Health Statistics data, 1979 to 1998. J Neurosurg 103:2 Suppl1131182005

4

Colwell RRHuq AIslam MSAziz KMAYunus MKhan NH: Reduction of cholera in Bangladeshi villages by simple filtration. Proc Natl Acad Sci U S A 100:105110552003

5

Corkeron MLNorton RNelson PN: Spatial analysis of melioidosis distribution in a suburban area. Epidemiol Infect 138:134613522010

6

Currie BJJacups SP: Intensity of rainfall and severity of melioidosis, Australia. Emerg Infect Dis 9:153815422003

7

Ferraro RRGrody NCWeng FBasist A: An eight-year time series of rainfall, clouds, water vapor, snow cover, and sea ice derived from SSM/I measurements. Bull Am Meteorol Soc 77:8919061996

8

Hashizume MChaves LFMinakawa N: Indian Ocean Dipole drives malaria resurgence in East African highlands. Sci Rep 2:2692012

9

Hashizume MTerao TMinakawa N: The Indian Ocean Dipole and malaria risk in the highlands of western Kenya. Proc Natl Acad Sci U S A 106:185718622009

10

Hassan MRPani SPPeng NPVoralu KVijayalakshmi NMehanderkar R: Incidence, risk factors and clinical epidemiology of melioidosis: a complex socio-ecological emerging infectious disease in the Alor Setar region of Kedah, Malaysia. BMC Infect Dis 10:3022010

11

Herman AKumar VBArkin PAKousky JV: Objectively determined 10-day African rainfall estimates created for famine early warning systems. Int J Remote Sens 18:214721591997

12

Holmes PLumley JLBerkooz G: CambridgeCambridge University Press1996

13

Koelle KRodó XPascual MYunus MMostafa G: Refractory periods and climate forcing in cholera dynamics. Nature 436:6967002005

14

Li LPadhi ARanjeva SLDonaldson SCWarf BCMugamba J: Association of bacteria with hydrocephalus in Ugandan infants. Clinical article. J Neurosurg Pediatr 7:73872011

15

Mitra PBokil H: Observed Brain Dynamics New YorkOxford University Press2008

16

Mitra PPPesaran B: Analysis of dynamic brain imaging data. Biophys J 76:6917081999

17

Netoff TICarroll TLPecora LMSchiff SJDetecting coupling in the presence of noise and nonlinearity. Schelter BWinterhalder MTimmer J: Handbook of Time Series Analysis WeinheimWiley-VCH Verlag GmbH & Co. KGaA2006. 265282

18

Roberts L: Infectious disease. An ill wind, bringing meningitis. Science 320:171017152008

19

Ruiz-Moreno DPascual MBouma MDobson ACash B: Cholera seasonality in madras (1901–1940): dual role for rainfall in endemic and epidemic regions. EcoHealth 4:52622007

20

Ruiz-Moreno DPascual MEmch MYunus M: Spatial clustering in the spatio-temporal dynamics of endemic cholera. BMC Infect Dis 10:512010

21

Schiff SJ: Neural Control Engineering CambridgeMIT Press2012

22

Schiff SJHuang XWu JY: Dynamical evolution of spatiotemporal patterns in mammalian middle cortex. Phys Rev Lett 98:1781022007

23

Schiff SJSauer TKumar RWeinstein SL: Neuronal spatiotemporal pattern discrimination: the dynamical evolution of seizures. Neuroimage 28:104310552005

24

Schmidt E: Zur Theorie der linearen und nichtlinearen Integralgleichungen. I. Teil: Entwicklung willkürlicher Funktionen nach Systemen vorgeschriebener. Math Ann 63:4334761907

25

Tass PRosenblum MGWeule JKurths JPikovsky AVolkmann J: Detection of n:m phase locking from noisy data: application to magnetoencephalography. Phys Rev Lett 81:329132941998

26

Thomas ADForbes-Faulkner JParker M: Isolation of Pseudomonas pseudomallei from clay layers at defined depths. Am J Epidemiol 110:5155211979

27

Warf BC: Hydrocephalus in Uganda: the predominance of infectious origin and primary management with endoscopic third ventriculostomy. J Neurosurg 102:1 Suppl1152005

28

Warf BCAlkire BCBhai SHughes CSchiff SJVincent J: Infant hydrocephalus in sub-Saharan Africa: Cost and benefit of neurosurgical intervention. Clinical article. J Neurosurg Pediatr 8:5095212011. (Erratum in J Neurosurg Pediatr 9:

TrendMD

Metrics

Metrics

All Time Past Year Past 30 Days
Abstract Views 4 4 4
Full Text Views 19 19 14
PDF Downloads 12 12 9
EPUB Downloads 0 0 0

PubMed

Google Scholar