Relationships between climate and year-to-year variability in meningitis outbreaks: A case study in Burkina Faso and Niger
- Pascal Yaka^{1},
- Benjamin Sultan^{2}Email author,
- Hélène Broutin^{3, 4},
- Serge Janicot^{2},
- Solenne Philippon^{1} and
- Nicole Fourquet^{1}
https://doi.org/10.1186/1476-072X-7-34
© Yaka et al; licensee BioMed Central Ltd. 2008
Received: 24 January 2008
Accepted: 02 July 2008
Published: 02 July 2008
Abstract
Background
Every year, West Africa is afflicted with Meningococcal Meningitis (MCM) disease outbreaks. Although the seasonal and spatial patterns of disease cases have been shown to be linked to climate, the mechanisms responsible for these patterns are still not well identified.
Results
A statistical analysis of annual incidence of MCM and climatic variables has been performed to highlight the relationships between climate and MCM for two highly afflicted countries: Niger and Burkina Faso. We found that disease resurgence in Niger and in Burkina Faso is likely to be partly controlled by the winter climate through enhanced Harmattan winds. Statistical models based only on climate indexes work well in Niger showing that 25% of the disease variance from year-to-year in this country can be explained by the winter climate but fail to represent accurately the disease dynamics in Burkina Faso.
Conclusion
This study is an exploratory attempt to predict meningitis incidence by using only climate information. Although it points out significant statistical results it also stresses the difficulty of relating climate to interannual variability in meningitis outbreaks.
Background
The actual factors that initiate these epidemics are not yet clearly understood although we know that a complex interplay of social interactions [6], transmission of a new epidemic strain [4], susceptibility of populations [7], asymptomatic carriage [8] and environmental conditions [9] is involved. Among favourable conditions for the resurgence and dispersion of the disease, climatic conditions may be important inducing seasonal fluctuations in disease incidence and contributing to explain the spatial pattern of the disease roughly circumscribed to the ecological Sahelo-Sudanian band [2, 10]. The role of climate on this meningitis seasonality and spatial distribution has been widely studied [11–13]. The Sahelo-Sudanian region is subjected to a sequence of dry winter, dominated by dry and dust-laden northern winds, called the Harmattan, and wet season starting at spring with the monsoon. A recent study has provided a clear, quantitative demonstration of the existing connections between meningococcal meningitis epidemics onset and the Harmattan winds [11]. The authors have shown a correlation between the week of the maximum speed of a Harmattan wind index and the week of the onset of the epidemics in Mali. Another recent study [14] found that anomalies in annual meningitis incidence at district level were related to monthly climate anomalies. Significant relationships were found for both estimates of dust and rainfall in the pre-, post- and epidemic season.
The objective of this study is twofold. First, we investigate the role of climate on the triggering of MCM epidemics by using a long-term dataset. Second, we explore the possibility to include the climate conditions as a predictor of meningitis epidemics. To do so, we start by defining an a priori hypothesis on the causal link between climate and disease. Based on a literature review, we assume that dry and windy weather conditions in early winter might cause damage to the mucous membranes of the respiratory system and/or inhibits mucosal immune facilitating the transfer of the bacterium to the meninges and thus create propitious conditions to the triggering of MCM epidemics [12, 13, 15]. According to this hypothesis, if the role of climate is strong enough, we should observe a positive correlation between the markers of these particular winter conditions (i.e. strong north-easterly wind, high pressure and dryness) and the MCM incidence. Since the triggering of epidemics is probably not only due to climate but results from numerous processes acting at different spatial hierarchical scales in various medical, demographical and socio-economical conditions, an absence of significant correlation between climate and disease would not necessary confound this hypothesis but could point out that climate is not a major driver. In order to verify our a priori hypothesis, we will examine the statistical links between winter climate variables and disease dynamics in two highly affected countries, Niger and Burkina Faso. Since most studies have focused on very small spatial scales, we will follow suggestions from Sultan et al. [11] and Broutin et al. [3] and use national-scale aggregated data. The aggregation of local data is a simple way to go beyond data heterogeneities and idiosyncratic details in order that only the important disease generalities, conditioned by the large-scale forcing, e.g. climate variability, remain.
Materials and methods
Epidemiological data
This work is based on the WHO disease surveillance over Africa. Details on the diagnosis used for meningitis can be found in WHO [16]. Meningitis reporting is incorporated into weekly reporting of notifiable diseases and aggregated at different spatial scales from the health unit to the country level. From this surveillance, WHO proposed a strategy aimed at early detection and control of meningitis epidemics at the level of health districts. This strategy is based on a strengthened epidemiological surveillance, mass immunization campaigns when incidence rate thresholds are exceeded and case management with appropriate antimicrobial therapy.
This database is available online on the WHO website and has already been used by Broutin et al. [3] to make a comparative study of meningitis dynamics across nine African countries. Although this dataset covers the period 1939–2005, we consider only the years after 1966 for two reasons. First, there is a large number of missing values before 1966. Second, Broutin et al. [3] have shown major changes after the 60's and the 70's in meningitis periodicities and synchronicities across African countries. These changes might be induced by the start of vaccinations at the end of 1970's in all of these countries. By taking into account only the post-vaccination period, we consider the most homogeneous time series to describe the meningitis dynamics but we have to keep in mind that a part of the variance of the incidence data might be explained by vaccination effect. However, as vaccination strategy has varied across country and within a same country (different proportions and with different types of vaccines, different policy from reactive to mass vaccination), the impact of vaccination is very difficult to point out in our time series.
Summary of epidemiological and environmental datasets
Type of data | Available data period | Used data period | Available time/space scale | Used time/space scale | |
---|---|---|---|---|---|
1 | Meningitis cases | 1939-today | 1966–2005 | Weekly/Health unit | Annual/Country |
2 | Environmental variables | 1948-today | 1968–2005 | 4 values per day/grid-spacing of 2.5° latitude by 2.5° longitude | Monthly/Country |
Classification of the MCM datasets for 8 West African countries
Very low IR < 7.2 | Low 7.2 < IR < 18.9 | High 18.9 < IR < 41.3 | Very high IR > 41.3 | ||
---|---|---|---|---|---|
1 | Benin | 4 | 15 | 9 | 5 |
2 | Burkina Faso | 0 | 1 | 18 | 15 |
3 | Chad | 5 | 9 | 9 | 11 |
4 | Mali | 10 | 12 | 4 | 6 |
5 | Niger | 0 | 1 | 9 | 20 |
6 | Nigeria | 19 | 9 | 2 | 1 |
7 | Sudan | 18 | 2 | 6 | 4 |
8 | Togo | 8 | 14 | 7 | 2 |
Atmospheric data
The National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) have completed a reanalysis project with an up-to-date version of the Medium Range Forecast model [17]. This dataset consists in a reanalysis of the global observational network of meteorological variables and a forecast system to perform data assimilation throughout the period 1948 to present. However, prior to 1968 they are not fully reliable for the African continent, as demonstrated in Camberlin et al. [18].
The environmental variables used in the study
Environmental variable | Code | |
---|---|---|
1 | Zonal wind (m/s) | UWND |
2 | Meridional wind (m/s) | VWND |
3 | Wind speed (m/s) | MOD |
4 | Sea level pressure (Pa) | SLP |
5 | Surface temperature (°C) | AIR |
6 | Surface relative humidity (%) | RHUM |
7 | Surface specific humidity (kg/kg) | SHUM |
- Wind velocity (zonal and meridional components, wind speed) and sea-level pressure that characterize the Harmattan circulation intensity. The influence of the latter circulation on meningitis has been shown recently [11].
- Surface temperature, specific and relative humidity near the surface that are markers of the dry conditions propitious to the MCM epidemics. The incidence of MCM has previously been correlated with dry and dusty conditions [1, 9, 13].
Correspondence Analysis
In order to summarize the MCM cases dataset composed of 34 annual values from 1966 to 1999 for each of the 8 West African countries (Table 1 and Table 2) under study and to compare the disease dynamics in these countries, we use the Correspondence Analysis (CA). This method is well-suited to look at the main structure of a normally distributed dataset without a priori hypothesis. Since we normalize the distribution of meningitis data by applying a log-transform to IR, this method can thus be applied to our data. For that:
- First, since the log-IR are normally distributed, it permits classification of the years according to quartile groups which divide the sorted data set into four equal parts so that each part represents 1/4 of the sampled population. For each country, we then classified the years of meningitis data into one of the four categories according to log-IR thresholds. The first 1/4 is defined as very low incidence group, the second is the "low incidence" group, the third is "high incidence" and the last one is the "very high" (Table 2). As a result, for Benin for instance, 4 years are considered with "very low" incidence, 15 years are "low", 9 are "high" and 5 are "very high".
- The IR ranges of the four resulting categories (Table 2) are pared to the thresholds defined by WHO for the epidemics surveillance. Two thresholds are used by WHO at the district level, the Alert Threshold (AT) which considers more than 5 cases per week and per 100.000 inhabitants and the Epidemic Threshold (ET) which is more than 10 cases per week and per 100.000 inhabitants. The lowest IR category regroups the years with less than 7.2 cases/100.000 inhabitants which is lower than the ET while the highest IR category regroups years with an IR four times greater than the ET. However, since data are used at a more aggregated space and time scale (country instead of district and year instead of week), there is no direct correspondence between these thresholds and the quartiles.
- We then construct a two-way contingency table from the original dataset which is a tabular cross-classification of data such that one subcategory (countries) is indicated in rows and another subcategory (the four IR categories) is indicated in columns (Table 2).
- Finally, we apply a CA which is able to give the best simultaneous representation of the rows and the columns of such two-way contingency table. CA is based on the extraction of the principal canonical correlations and corresponding row and column-scores from a correspondence analysis of a two-way contingency table [19]. From the row and column-scores on the two main factors, it is possible to make a graphical representation of both rows (countries) and columns (the four IR categories) of the contingency table (see Fig. 2). Such graphical representation gives a good summary of the structure of the original dataset highlighting proximities between one country and another and between countries and IR categories. Notice that the distance between two countries is not driven by the average IR over the period 1966–1999 but by the number of years in each cluster. We thus are able to compare the distribution of MCM annual IR for each country.
Composite analysis
In order to detect relationships between climate and MCM incidence, we perform a composite analysis. For each country, we first classify the years into two sub-groups: the years with the highest IR (the years with a log-IR greater than the third quartile) and the years with the lowest IR (the years with a log-IR lower than the first quartile). We then average separately the atmospheric data for the high (HIGH) and low (LOW) MCM incidence years. Finally we compute the HIGH minus LOW difference in order to point out the atmospheric situations characterizing a typical high incidence year. The significance of this difference is attested using a Student test.
Correlations between climate and MCM
In order to detect relationships between climate and MCM incidence, we compute the correlations between the monthly atmospheric variables and the annual log-IR. For each country, we compute the correlation coefficients between each of the 7 atmospheric values for one month and the annual log-IR of the country. These correlations are given for the 4 fall-winter months: October to January. They are computed over 1968–2005. Because of the number of variables used for the correlations computation (7 variables over 4 months), there is always the possibility of a chance association emerging. Classical significance tests do not consider this risk since there are applied independently for each time series and do not take into account the repetition of the correlation computation which increases the possibility of a chance association. We thus set-up a test reproducing our experimental conditions: we generate a set of 7 × 4 random Gaussian time series with the same properties (mean and variance) than the atmospheric variables and compute the correlation between each random time series and the log-IR of the considered country. We then use to assess the significance of the correlations at the 1% level of confidence the threshold given by the positive (negative) correlation values of the 99% (1%) quantile of the 7 × 4 random correlation values. This experiment is reproduced 10, 000 times and we average the positive and negative thresholds over the 10, 000 repetitions.
Multivariate linear regression models
The analysis of the links between climate and MCM are then used to select a set of atmospheric predictors that likely influence the MCM incidence. These predictors can be used to build up a stepwise multivariate linear regression model in order to predict the annual MCM incidence rate. The equation of the multivariate model is:
Y = α + β_{1}X_{1} + ⋯ + β_{ p }X_{ p }
Y represents the predicted log-IR value, α is a constant, and each β terms denotes a regression coefficient for the corresponding predictor X.
The robustness and the forecast skill of the regression models are assessed using two standard methods: the cross-validated correlation and the Relative Operating Characteristics (ROC) score:
- A simple leave-one-out cross-validation is used to document the stability of the regression models: we compute the model parameters from a portion of the data (called training period) composed by all years minus one and we look at the prediction of the remaining data. The cross-validated correlation is a much more realistic representation of the skill of the model applied to "unseen" years.
- ROC is a means of testing the skill of categorical forecasts [20, 21]. It is based on contingency tables giving the Hit Rate (HR) and False Alarm Rate (FAR). We first transform our data and forecasts into binary time series where only two outcomes are possible, an occurrence of high incidence year or a non-occurrence, according if the log-IR of the year is greater or above the median which divides the dataset in half. We then compute a contingency table based on these two categories and calculate the HR and FAR which are simply percentages that tell us how well the forecast did when a high incidence year was observed, and likewise, how well the forecast did when a high incidence year was not observed. An example of this contingency table is given in Table 4. The "hits" ("zeros") category represents the number of high (non-high) incidence rate that have been forecasted as so. The "false alarms" ("misses") category represents the number of non-high (high) incidence years that have been forecasted as high (non-high) incidence years. The HR is defined as:
Hits and False alarms in forecast models
Predicted | |||
---|---|---|---|
Non-High incidence year | High incidence year | ||
Observed | Non-High incidence year | zeros | false alarms |
High incidence year | Misses | hits |
The range of possible values goes from 0 to 1 where a perfect forecast system has a value of one and a forecast system with no information has an value of 0.5 (HR being equal to FAR).
In our study, these standard forecast skill methods are also applied to a reference forecast method based on persistence (the incidence rate of one year is the same than the incidence rate of the previous year). Since the persistence is the simplest way to produce a forecast, we consider the skill of our regression models as useful if it is greater than the persistence skill.
Susceptible-Infected Model
In order to illustrate hypotheses on the disease transmission, we use a Susceptible-Infected (SI) model. Such models [22] are widely used for direct infectious disease in order to examine transmission processes [23]. It consists of two compartments: Susceptible (S) and Infected (I). Individuals in the S compartment are susceptible to be infected and move to the I compartment with a speed controlled by a transmission rate. The initial sizes of the two compartments and the transmission rate are the parameters needed to fit the model. We choose these parameters to reproduce roughly the same temporal characteristics of an outbreak in Burkina Faso. To do so, we use a discrete model and simulate an outbreak over one virtual year of 52 time steps (a weekly time step) with a peak occurring during the first half of the year. Notice that the SI model is mainly used here for an illustrative purpose.
Results
The year-to-year variability of meningitis cases and incidence rates is described for 8 West African countries over the 34-year period between 1966 and 1999. The CA is used to compare the disease dynamics in these countries (see Materials and Methods). Figure 2 represents graphically the 8 countries in the two main factors of the CA applied to the MCM incidence rates. The proximity between countries is driven by the number of years in the 4 incidence categories (Table 2). The main factor (the horizontal axis) discriminates (i) the countries with a high number of low MCM incidence years (i.e. Sudan, Nigeria) on the left side of the axis and (ii) the highly affected countries with the greatest number of high MCM incidence years (Burkina Faso and Niger) on the right side of the axis. The second factor (the vertical axis) isolates the countries with a large number of years whose annual MCM incidence rates are close to the 1966–1999 average (high and low, i.e. Mali, Togo, Chad and Benin). The analysis thus points out three different disease dynamics in these countries: the "high-risk" countries characterized by several years with very severe epidemics, the "medium-risk" countries mainly with MCM cases each year but without severe epidemics, the "low-risk" countries with very low MCM cases each year. We now focus our study on the two "high-risk" countries, i.e. Niger and Burkina Faso to analyze the link between MCM incidence and climate.
Skill scores of the meningitis forecast models
Niger | Burkina Faso | |||
---|---|---|---|---|
Climate-based model | Persistence model | Climate-based model | Persistence model | |
CVC | 0.50 | 0.30 | 0.33 | 0.55 |
HR | 0.83 | 0.67 | 0.74 | 0.79 |
FAR | 0.47 | 0.41 | 0.50 | 0.28 |
HKS | 0.68 | 0.63 | 0.62 | 0.76 |
The Burkina Faso model gives lower scores with a correlation between the observed and predicted MCM time series around 0.42 and falling to 0.33 for the CVC. Forecasts based on persistence do a better job for the four skill scores CVC, HR, FAR and HKS (see Table 5). These low skill scores can be partly explained by the fact that only one predictor in the atmospheric circulation could be found to build the regression model while two predictors were used in Niger.
Discussion
In this study, the relationships between climate and MCM disease at interannual and country scales have been statistically investigated in two highly affected countries: Niger and Burkina Faso. We pointed that these links are particularly clear in Niger and weak but significant in Burkina Faso. The disease resurgences in Niger and in Burkina Faso are linked with an enhancement of the winter conditions, e.g. enhanced Harmattan winds over Niger in November/December and over Burkina Faso in October. These findings are coherent with a previous study which showed a positive correlation between the October dust and meningitis incidence in Burkina Faso, Mali and Niger [14]. Here, we also defined relevant climatic variables for the construction of linear models to forecast MCM epidemics intensity from year to year. These statistical models work well for Niger showing that 25% of the disease variance from year-to-year in this country can be explained by the winter climate but fail to represent accurately the disease dynamics in Burkina Faso. Although this study points out significant statistical results, it also stresses the difficulty of relating climate to interannual variability in meningitis outbreaks. Numerous reasons can be pointed out to explain this limitation but two are of most importance:
First, the final size of the outbreak clearly does not depend only on climate but implies many other factors. The size of the epidemics will be also (and perhaps mainly) driven by the immunity of the affected population against the serotype involved in the outbreak and socio-economic factors (pilgrimages, migrations) [6, 7]. The proportion of carriers might also play an important role in the disease dynamics (see [8] for review). Vaccination, even if there is a debate on the efficiency on the meningitis control activities [24–26], has certainly an impact on the final size of the outbreak as it is suggested by Broutin et al. [3] in their comparative studies of meningitis dynamics across several African countries. The second important limitation to our study corresponds to the meningitis data themselves. Missing values as well as suspected underreporting may introduce some biases in the incidence time series. As a consequence of the above factors, influencing the final size of the outbreak, it is thus very likely that the meningitis incidence data contain trends, strong or low incidence events or periods that can not be related to any climate effect. It is probably what starts to account for the differences in results between countries. The weak correlation between climate and disease in Burkina Faso does not necessary confound our hypothesis of the dry northerly winds being implicated in the outbreaks of meningitis but could point out that climate is not a major driver of the disease dynamics. Alternatively, since the variability of the meningitis incidence from one year to another results from numerous processes acting at different spatial hierarchical scales in various medical, demographical and socio-economical conditions, the results obtained with the Niger model, even if the correlation values and the explained variance are weak, suggest that climate is an important driver for the triggering of epidemics in Niger and validate the large-scale approach that allows to smooth local data heterogeneities.
Given these difficulties, much more work need to be done to use such climate indexes in the context of a survey and an early warning system (EWS) to influence public health policy, for example setting in place an epidemic preparedness programme, ordering vaccine, etc. After this work, it is clear that the opportunity of the use of environmental variables in such EWS should now be further investigated by including other factors which also are likely important drivers in MCM dynamics. In particular, epidemiological and population parameters (immunity, carriers, population size, ...) should clearly be including in any forecasting tool as well as behaviour factors like migrations. This approach will remain explorative since the main risk and control factors for the disease and how they interplay each other are not better understood. Further combined epidemiological and climate studies are recommended to help in a better understanding of MCM dynamics and evolution at different spatial-scales. A key issue is the extent and the improvement of epidemiological and environmental datasets through long-term longitudinal studies (see for instance [7] for the immunological factors) and collection of both environmental and epidemiological data over the same site [29]. These improvements are necessary but might not be sufficient to provide highly precise forecasts since stochastic processes in the transmission could limit the predictability of the final size of the outbreak. The limits to the precision of EWS for epidemics of infectious disease have been discussed recently by Drake [23]. According to the author, the characteristics of emerging diseases to which human populations are highly susceptible prevent precise forecasts because of the micro-scale component (contacts within and among households and communities) whose small variations could induce large variations in the final size of the outbreak. While the forecast of the EWS based on climate and other environmental characteristics contain only the macro-scale source of variation and not the micro-scale causes, they can still be used effectively to define risk indicators rather than precise forecasts that could be used to better control MCM disease.
Declarations
Acknowledgements
We are thankful to J-P Chippaux, S. Kennan and the four anonymous reviewers who helped to clarify this paper. BS and SJ thank IRD for financial support.
Authors’ Affiliations
References
- Greenwood B: Manson Lecture. Meningococcal meningitis in Africa. Trans R Soc Trop Med Hyg. 1999, 93: 341-353. 10.1016/S0035-9203(99)90106-2.PubMedView ArticleGoogle Scholar
- Lapeyssonnie L: The Meningococcal meningitis in Africa (in french). Bull WHO. 1963, 28–3: 114-Google Scholar
- Broutin H, Philippon S, Constantin de Magny G, Courel F, Sultan B, Guegan JF: Comparative study of meningitis dynamics across nine African countries: a global perspective. Int J Health Geo. 2007, 6: 29-10.1186/1476-072X-6-29. doi:10.1186/1476-072X-6-29.View ArticleGoogle Scholar
- Greenwood B: 100 years of epidemic meningitis in West Africa – has anything changed?. Trop Med Int Health. 2006, 11: 773-780. 10.1111/j.1365-3156.2006.01639.x.PubMedView ArticleGoogle Scholar
- Teyssou R, Muros-Le Rouzic E: Meningitis epidemics in Africa: A brief overview. Vaccine. 2007, 25: A3-A7. 10.1016/j.vaccine.2007.04.032.PubMedView ArticleGoogle Scholar
- Hodgson A, Smith T, Gagneux S, Adjuik M, Pluschke G, Kumasenu Mensah N, Binka F, Genton B: Risk factors for meningococcal meningitis in northern Ghana. Trans R Soc Trop Med Hyg. 2001, 95: 477-480. 10.1016/S0035-9203(01)90007-0.PubMedView ArticleGoogle Scholar
- Leimkugel J, Hodgson A, Forgor AA, Pflüger V, Dangy JP, Smith T, Achtman M, Gagneux S, Pluschke G: Clonal waves of Neisseria colonisation and disease in the African meningitis belt: eight- year longitudinal study in northern Ghana. PLoS Med. 2007, 4: e101-10.1371/journal.pmed.0040101.PubMedPubMed CentralView ArticleGoogle Scholar
- Trotter CL, Greenwood BM: Meningococcal carriage in the African meningitis belt. Lancet Infect Dis. 2007, 7: 797-803. 10.1016/S1473-3099(07)70288-8.PubMedView ArticleGoogle Scholar
- Molesworth AM, Cuevas LE, Connor SJ, Morse AP, Thomson MC: Environmental risk and meningitis epidemics in Africa. Emerging Infectious Disease. 2003, 9: 1287-1293.View ArticleGoogle Scholar
- Cheesbrough JS, Morse AP, Green SDR: Meningococcal meningitis and carriage in western Zaire : A hypoendemic zone related to climate?. Epidemiology and Infectious. 1995, 114: 75-92.View ArticleGoogle Scholar
- Sultan B, Labadi K, Guegan JF, Janicot S: Climate drives the meningitis epidemics onset in west Africa. PLoS Med. 2005, 2: e6-10.1371/journal.pmed.0020006.PubMedPubMed CentralView ArticleGoogle Scholar
- de Chabalier F, Djingarey MH, Hassane A, Chippaux JP: Meningitis seasonal pattern in Africa and detection of epidemics: a retrospective study in Niger, 1990–98. Trans R Soc Trop Med Hyg. 2000, 94: 664-668. 10.1016/S0035-9203(00)90224-4.PubMedView ArticleGoogle Scholar
- Greenwood BM, Blakebrough IS, Bradley AK, Wali S, Whittle HC: Meningococcal disease and season in sub-Saharan Africa. Lancet. 1984, 1: 1339-1342. 10.1016/S0140-6736(84)91830-0.PubMedView ArticleGoogle Scholar
- Thomson MC, Molesworth AM, Djingarey MH, Yameogo KR, Belanger F, Cuevas LE: Potential of environmental models to predict meningitis epidemics in Africa. Trop Med Int Health. 2006, 11: 781-788. 10.1111/j.1365-3156.2006.01630.x.PubMedView ArticleGoogle Scholar
- Moore PS: Meningococcal meningitis in sub-Saharan Africa: a model for the epidemic process. Clin Infect Dis. 1992, 14: 515-25.PubMedView ArticleGoogle Scholar
- WHO: Guidelines on the Control of Epidemic Meningococcal Disease, WHO/EMC/BAC/98.3. 1998, secondGoogle Scholar
- Kalnay E: The NCEP/NCAR 40-year reanalysis project. Bull Amer Meteorol Soc. 1996, 77: 437-471. 10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.View ArticleGoogle Scholar
- Camberlin P, Janicot S, Poccard I: Seasonality and atmospheric dynamics of the teleconnections between African rainfall and tropical ocean surface temperature: Atlantic vs. ENSO. Int J Climatol. 2001, 21: 973-1005. 10.1002/joc.673.View ArticleGoogle Scholar
- Venables WN, Ripley BD: Modern Applied Statistics with S-PLUS. 2002, New-York, Springer-Verlag, 4View ArticleGoogle Scholar
- Mason SJ, Graham NE: Conditional Probabilities, Relative Operating Characteristics, and Relative Operating Levels. Weather and Forecasting. 1999, 14: 713-725. 10.1175/1520-0434(1999)014<0713:CPROCA>2.0.CO;2.View ArticleGoogle Scholar
- WMO: Standardised Verification System for Long-Range Forecasts, New Attachment II-8 to the Manual on the GDPFS (WMO-No. 485). 2006, I:Google Scholar
- Kermack WO, McKendrick AG: A contribution to the mathematical theory of epidemics. Proc R Soc Lond A. 1927, 115: 700-721. 10.1098/rspa.1927.0118.View ArticleGoogle Scholar
- Drake JM: Limits to forecasting precision for outbreaks of directly transmitted diseases. PLoS Med. 2006, 3: e3-10.1371/journal.pmed.0030003.PubMedPubMed CentralView ArticleGoogle Scholar
- Robbins JB: Meningococcal meningitis in sub-Saharan Africa: the case for mass and routine vaccination with available polysaccharide vaccines. Bull WHO. 2003Google Scholar
- Robbins JB, Towne DW, Gotschlich EC, Schneerson R: "Love's labours lost": failure to implement mass vaccination against group A meningococcal meningitis in sub-Saharan Africa. Lancet. 1997, 350: 880-882. 10.1016/S0140-6736(97)03227-3.PubMedView ArticleGoogle Scholar
- Chippaux JP, Debois H, Saliou P: A critical review of control strategies against meningococcal meningitis epidemics in sub-Saharan African countries. Infection. 2002, 30: 216-224. 10.1007/s15010-002-3012-2.PubMedView ArticleGoogle Scholar
- Tikhomirov E, Santamaria M, Esteves K: Meningococcal disease: public health burden and control. Rapp Trimest Stat Sant Mond. 1997, 50: 170-177.Google Scholar
- WHO: Detecting meningococcal meningitis epidemics in highly-endemic African countries. Weekly Epidemiological Record. 2000, 38:Google Scholar
- Cuevas LE, Jeanne I, Molesworth A, Bell M, Savory EC, Connor SJ, Thomson MC: Risk mapping and early warning systems for the control of meningitis in Africa. Vaccine. 2007, 25: A12-A17. 10.1016/j.vaccine.2007.04.034.PubMedView ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.