Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 a B D U S S a L a M E T a L

62 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

Climate Inﬂuences on Meningitis Incidence in Northwest Nigeria

AUWAL F. ABDUSSALAM School of Geography, Earth, and Environmental Sciences, University of Birmingham, Birmingham, United Kingdom

ANDREW J. MONAGHAN National Center for Atmospheric Research,* Boulder, Colorado

VANJA M. DUKIC Department of Applied Mathematics, University of Colorado, Boulder, Boulder, Colorado

MARY H. HAYDEN AND THOMAS M. HOPSON National Center for Atmospheric Research,* Boulder, Colorado

GREGOR C. LECKEBUSCH AND JOHN E. THORNES School of Geography, Earth, and Environmental Sciences, University of Birmingham, Birmingham, United Kingdom

(Manuscript received 18 December 2012, in ﬁnal form 6 August 2013)

ABSTRACT

Northwest Nigeria is a region with a high risk of meningitis. In this study, the influence of climate on monthly meningitis incidence was examined. Monthly counts of clinically diagnosed hospital-reported cases of meningitis were collected from three hospitals in northwest Nigeria for the 22-yr period spanning 1990– 2011. Generalized additive models and generalized linear models were fitted to aggregated monthly meningitis counts. Explanatory variables included monthly time series of maximum and minimum temperature, humidity, rainfall, wind speed, sunshine, and dustiness from weather stations nearest to the hospitals, and the number of cases in the previous month. The effects of other unobserved seasonally varying climatic and nonclimatic risk factors that may be related to the disease were collectively accounted for as a flexible monthly varying smooth function of time in the generalized additive models, s(t). Results reveal that the most important explanatory climatic variables are the monthly means of daily maximum temperature, relative humidity, and sunshine with no lag; and dustiness with a 1-month lag. Accounting for s(t) in the generalized additive models explains more of the monthly variability of meningitis compared to those generalized linear models that do not account for the unobserved factors that s(t) represents. The skill score statistics of a model version with all explanatory variables lagged by 1 month suggest the potential to predict meningitis cases in northwest Nigeria up to a month in advance to aid decision makers.

1. Introduction * The National Center for Atmospheric Research is sponsored by the National Science Foundation. The strictly human pathogen Neisseria meningitidis is a gram-negative diplococcus that mainly causes meningitis (Rosenstein et al. 2001; Schuchat et al. 1997). De- Corresponding author address: Auwal F. Abdussalam, School of Geography, Earth, and Environmental Sciences, University of spite the existence of several bacteria that can cause Birmingham, Edgbaston, Birmingham B15 2TT, United Kingdom. meningitis, this bacterium is responsible for causing large E-mail: [email protected] epidemics [World Health Organization (WHO) 2012].

DOI: 10.1175/WCAS-D-13-00004.1

Ó 2014 American Meteorological Society Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 63

The associated symptoms of meningitis may include An extensive part of Nigeria lies within the meningitis headache, fever, vomiting, and a stiff neck. The disease is belt (see appendix A). The disease primarily affects the a major public health burden in several countries around country during the dry months, beginning with the the world (Peltola 1983), but its magnitude is more pro- Harmattan (a dry and dusty northerly wind) in November found in Africa, in an area mainly in the Sahelian region and continuing through May with peak incidence during recognized as the meningitis belt by Lapeyssonnie (1963) the hottest months of March and April. During the 1996 and acknowledged by many (e.g., Greenwood 2006; epidemic, Nigeria alone reported over 100 000 cases and Harrison et al. 2009; Molesworth et al. 2002). 11 000 deaths to the World Health Organization (WHO) Although the mechanisms by which climatic factors (Mohammed et al. 2000), which is almost half of the total influence meningitis are still not well understood, it is cases reported from the 25 countries within the belt. Most believed that increased concentrations of dust, high of these cases and deaths were reported from the north- winds, elevated temperatures, and low humidity may ern part of the country, where there is a more pronounced cause damage to the nasopharyngeal mucosa (WHO dry season than in the south, and where the disease mainly 2012), thereby increasing meningitis risk. Outbreaks occurs each year. Meningitis in northern Nigeria is known have been associated with climatic and environmental in the native Hausa language as ‘‘Sankarau,’’ literally factors in the meningitis belt (e.g., Besancenot et al. meaning ‘‘the disease of stiffness,’’ obviously named be- 1997; Mueller et al. 2008; Sultan 2005; Thomson et al. cause of the stiff neck often associated with the disease, 2006; Yaka et al. 2008). The annual peak incidence of caused by inflammation of tissues that surround the brain meningitis has been found to correlate significantly with and the spinal cord. Anecdotal evidence suggests that the highest mean temperatures, and to correlate in- many people in northern Nigeria believe that the disease versely with absolute humidity and rainfall (Dukic et al. is caused by the intense heat usually experienced during 2012). Although some studies have concluded that prev- outbreaks. In Nigeria, meningitis transmission and in- alence of carriage does not vary with season (Greenwood vasion may be influenced by many factors apart from et al. 1984; Trotter and Greenwood 2007), others have climate, such as socioeconomic and cultural practices. In found a statistically significant link with seasonal and the cities, areas mostly occupied by low-income earners spatial climatic variability (Kristiansen et al. 2011). In are densely populated with many people per household, the Sahel, the hypothesis that meteorological factors and most use wood as their source of cooking fuel. Based influence the seasonality of disease transmission (e.g., on previous studies such as Hodgson et al. (2001), people Jandarov et al. 2012) and the development of invasive living in these areas are likely at higher risk and more disease is supported by the idea that the prolonged dry vulnerable to the development, carriage, and transmission season, which normally starts in the ‘‘cold dry’’ months of invasive meningitis. (November–January), may facilitate the occurrence of For the past 30 years, meningitis in Africa and Nigeria, precursory diseases that increase meningitis risk in the in particular, has been occurring in sporadic cases, out- subsequent ‘‘hot dry’’ months (e.g., Cartwright 1995). breaks, or even in large epidemics (Greenwood 2006) These previous efforts connecting climatic conditions to mainly caused by serogroup A (Greenwood and Stuart meningitis incidence suggest the possibility of predicting 2012; Riou et al. 1996). Until recently the only strategy meningitis outbreaks using environmental information. for controlling the disease has been through reactive mass Other nonclimatic factors may also play important vaccination campaigns after crossing a certain case roles in the risk of meningitis transmission. Among these threshold (WHO Working Group 1998), using poly- factors are socioeconomic, cultural, and behavioral saccharide A vaccine, and sometimes through special practices; and migration. Previous studies have estab- campaigns intended for religious pilgrims. Although the lished the association between nonclimatic factors and efficacy of this vaccine has been established, it is expen- meningitis risk—for example, previous incidences of sive and has a short period of effectiveness (Wahdan other upper respiratory tract infections (URTIs), such as et al. 1973). The introduction of the conjugate vaccine pneumococcal pneumonia (Moore et al. 1990), exposure (WHO 2012) brings hope for controlling the disease. to smoke from cooking fires (Hodgson et al. 2001), The new vaccine will not only provide longer protection overcrowding (Brundage and Zollinger 1987), disco but also prevent the disease (WHO 2012); that is, patronage (Cookson et al. 1998), and smoking (Fischer a whole population can be vaccinated proactively before et al. 1997). To attain precision in simulating and pos- meningitis epidemics are detected. Another advantage sibly predicting any kind of climate-related disease ep- of the conjugate vaccine is carriage prevention (Maiden idemic, if possible, climatic variables should be jointly et al. 2008; Vestrheim et al. 2010). Despite these advan- used with other nonclimatic factors as explanatory var- tages, since the conjugate vaccine is serogroup A specific iables (Thomson et al. 2006; Palmgren 2009). (Greenwood and Stuart 2012), other serogroups like

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 64 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

W135 (Decosas and Koama 2002) and C (Broome et al. a seasonal perspective, the variability of meteorological 1983) might continue to circulate, increasing the risk of conditions should relate to the variability of incidence in epidemics (Koumare et al. 1993). These other strains specific regions such as northwest Nigeria. could be introduced by travelers from outside regions b. Epidemiological data (Lingappa et al. 2003). The new conjugate A vaccine was recently adminis- Monthly counts of clinically diagnosed meningitis tered in Nigeria in 2011 (O. Kale et al. 2011, unpublished cases reported between 1990 and 2011 were collected manuscript). The vaccine was administered between 5 in situ from selected hospitals in the region; see Fig. 2. and 14 December 2011 during phase 1 (out of 3) of the The selection of hospitals was based on three criteria: campaign, which focused on the population group span- (i) proximity to meteorological stations with long-term ning 1–29 years of age in five northern Nigerian states: records of measurements, (ii) similar climatic patterns, Bauchi, Gombe, Jigawa, Katsina, and Zamfara. The cam- and (iii) consistently reported records of infectious dis- paign did not influence the results of this study because the ease cases. In Nigeria, the Federal Ministry of Health peak of our final year of meningitis case data had already (FMoH) classifies four categories of hospitals based on ceased before the administration of the vaccine. ownership status: federal, state, and local public hospi- The current study aims to statistically model the in- tals, and private hospitals. Personal communication fluence of climate on monthly meningitis incidence in with FMoH staff prior to the data collection indicated northwest Nigeria while collectively accounting for the that the state-owned hospitals best suited the above- effects of all unobserved climatic and nonclimatic fac- mentioned criteria because most of the infectious dis- tors that may be related to meningitis, such as social and ease cases are treated at these hospitals. Four hospitals behavioral practices, and migration. Our paper is the in major regional cities met the first two requirements: first to report a relationship between meteorological Kaduna, Kano, Sokoto, and Gusau. Kaduna was not conditions (reflected by variables from station obser- included in the model because it lies farther southward vations) and meningitis in Nigeria since the study by and has a distinct climatology, but the city was later used Greenwood et al. (1984); since that time, a much longer for testing the regional model. We also briefly sum- case record has been established. The model development marize the results of a separate model developed for and results are based on 22 years (1990–2011) of clinically Kaduna in the discussion. In addition to the monthly case diagnosed hospital-reported cases of meningitis. data from these three hospitals (1990–2011), we also obtained weekly records of suspected meningitis cases at the district level (from all hospitals in a district) in Nigeria 2. Materials and methods for the years 2007–11, from the WHO. The proportion of cases between WHO and hospital records was rather a. Study site and regional meteorological conditions constant across the study period, providing evidence that Northwest Nigeria is located in the African Sahel sa- the records are of sufficient quality and that the hospital vannah region (see appendix A), and currently has an records are regionally representative (see appendix B). estimated population of over 41 million people based on It is noteworthy, for the sake of interpreting the re- the 2006 census. The regional climate (Fig. 1) is char- sults, that the collected meningitis cases are suspected acterized by two seasons, a short wet season from June cases clinically diagnosed based on the WHO recom- to September and a prolonged dry season for the re- mended definition: as a person with sudden onset of mainder of the year. Daytime maximum temperatures fever (.38.58C rectal or 38.08C axillary) with one or remain consistently high throughout the year with more of following symptoms: neck stiffness, meningeal maxima during March–May (up to 478C), while relative sign, or altered consciousness (WHO 2010). humidity is low during the dry season and increases c. Meteorological data during the wet season. These mean regional climate conditions are mainly a consequence of the West Afri- Digital records of seven variables from airport-based can monsoon (WAM) system, which exhibits large meteorological stations in each of the three cities were spatiotemporal variability (e.g., Cornforth 2012), espe- obtained from the Nigerian Meteorological Agency cially with respect to regional rainfall distributions. The (Table 1). Daily precipitation and maximum and mini- WAM is a large-scale wind system characterized by mum temperatures were quality controlled using the moist northward flow from the Gulf of Guinea during software tool RClimDex.r (1.0), which has been de- the wet season and a dry and dusty southward flow veloped and maintained by the Climate Research Di- (Harmattan) during the dry season. As meningitis is vision (2008) of the Meteorological Service of Canada on known to decrease with the onset of the monsoon, from behalf of the Expert Team on Climate Change Detection

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 65

FIG. 1. Annual cycle of selected monthly meteorological variables, averaged over stations Kano, Sokoto, and Gusau in northwest Nigeria between 1990 and 2010. The vertical bars represent the 61 standard deviation from the monthly mean, and the gray shading represents the range of monthly means over the 21-yr record. and Indices. Other variables were manually quality con- Both lag zero and 1-month lagged meteorological data trolled by removing obvious spurious values based on were included as explanatory variables in the develop- expert knowledge of the regional climate; too few values ment of the models, while CO was only used for a sen- were removed to affect the overall quality and continuity sitivity test because the data spans too short a period for of the meteorological data. Monthly averages, totals, or inclusion in the final model development. percentages were then computed from the quality con- d. Demographic and vaccination campaign data trolled daily values. In addition to meteorological records, monthly to- District-level population census data as of 2006 were tal estimates of carbon monoxide (CO) emissions (Tg obtained from the National Population Commission in 2 month 1) for the subperiod from 1997 to 2009 were Abuja, Nigeria. Annual population estimates for each produced for northwest Nigeria by version 3.1 of the city were calculated forward and backward from 2006 Global Fire Emissions Database (GFEDv3.1) model using the Nigerian population growth rate index (World (van der Werf et al. 2010). These emissions estimates are Bank 2012). As vaccination campaigns will influence the driven by satellite observations of active fires, burn area, number of meningitis cases that would occur in the ab- and modeled vegetation information. sence of vaccination and thus can confound the model

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 66 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

FIG. 2. Map of Nigeria showing Kano, Sokoto, and Gusau, and the 10-yr mean of meningitis incidence rate per 100 000 of population between 2001 and 2011 for each state. State-level data were sourced from the Nigerian Centre for Disease Control (CDC) of the Federal Ministry of Health. results, we obtained information from FMoH on the Gusau were aggregated, and variables of the corre- specific years that reactive mass vaccination campaigns sponding three meteorological stations were averaged. were carried out in the selected cities. Using an estab- Distances between meteorological stations and the re- lished methodology described below, this information spective hospitals range between 3 and 6 mi. Monthly was used to estimate the expected number of meningitis meningitis counts were aggregated in order to minimize cases that would have occurred had there been no vac- the effect of bias in reporting to individual hospitals, and cination campaign. An additional model was then de- as well to have a regional perspective, which is the intent veloped using these expected cases as predictands, and of this study. Generalized additive model (GAM) and compared to the models that used the actual (unadjusted) generalized linear model (GLM) approaches were used. cases as predictands. GAMs are a flexible extension of GLMs and are com- prehensively described by Hastie and Tibshirani (1999). e. Model overview Because of the additive smoothing function within For statistical model development, monthly menin- the GAM (cf. below), we can collectively account gitis counts for the three hospitals in Kano, Sokoto, and (albeit not specifically) for the effects of all unobserved

TABLE 1. Summary of station characteristics and meteorological data at four stations in northwest Nigeria. Dustiness in this study means monthly percentage of the number of dry haze days (deﬁned as relative humidity below 80% and visibility below 5 km; WMO 1995) determined by recording visibility at a given observation point.

City Station Lat Lon Elev (m) Variables obtained from all stations Kano 65 0460 128030 088320 476 Max temp (8C), min temp (8C), rain (mm), 2 Sokoto 65 0100 128550 058120 351 relative humidity (%), wind speed (km h 1), dusty Gusau 65 0150 128100 068420 463 days (yes/no), and sunshine (h)

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 67 climatic and nonclimatic factors that may be related to and cases from the previous month; model B, with only the meningitis. 1-month lagged explanatory climatic variables and cases Both GAM and simpler GLM frameworks were from the previous month; and model C is the same as subsequently applied to model the influence of climatic GAM A, except previous cases were excluded. All three conditions on the monthly variability of clinically di- GAMs were tested for a variety of degrees of freedom agnosed hospital-reported cases of meningitis. During (DOF) for the fit of s(t), but it was found that those model development, collinearity diagnostics and auto- models in which s(t) has four DOF have the best fit. correlation checks were performed, and explanatory Model D, which is a GLM, has the same composition as variables were selected through a process of manually model A but without the smooth component s(t). In entering and removing variables from the model in a summary, model A is intended to be the optimal ex- stepwise selection process, with a criterion of elimina- planatory model of meningitis cases, whereas model B tion being a p value #0.05 when testing the significance by using only lagged variables is intended to be the op- of the coefficient estimate. The explanatory variables timal predictive model (with 1-month lead time). Model include meteorological variables aggregated monthly C, which does not include previous cases, is intended to between the three selected stations (Table 1), as well as be used for future climate change studies (since the the previous incidence of meningitis in some models. number of cases in the previous month is unknown in the Additionally, all other unobserved seasonally varying future). Model D is meant to test whether adding a climatic and nonclimatic factors that may influence the smooth function of time s(t) improves the model fit. All disease were represented in GAMs by a smooth function models were fit within R statistical software (R Core of time, s(t), which was modeled as a low-degree cubic Team 2009). spline that changes monthly over the course of the an- The best models were selected by minimizing the nual cycle (Dukic et al. 2012). The variable s(t) is the so- Bayesian information criteria (BIC). Variable selections called additive function characteristic of GAMs, and were made separately for each model, although the it captures the seasonality effect in a way that can be same variables were retained in all models but variables viewed as a smoothed analog of the month-specific ef- were all lagged by 1 month in model B, while previous fects. It is assumed that s(t) is common to all years (i.e., cases were not included in model C. The retained vari- that there is no intercept that can be applied to adjust the ables in models A and D include the current mean function for a specific year). We assume that the clini- monthly maximum and minimum temperatures; pre- cally diagnosed hospital meningitis counts yit follow in- cipitation totals; average relative humidity, sunshine, dependent Poisson distributions (thus, we use a log link and the 1-month lag of wind speed; dustiness; and cases function; Cameron and Trivedi 1998), with mean mit, (i.e., cases in the previous month). where i 5 1, ..., 22 denotes the years, and t 5 1, ...,12 A population offset term (to account for population denotes the month within each year. The GAM formu- growth) was not included in our models because esti- lation is thus mating the changes to the population served by a single hospital, the source of our records, is highly uncertain. m 5 1 X b log( it) s(t) it . For example, as population grows, new hospitals and other treatment facilities are built to accommodate more

The expected meningitis count [E(yit) 5 mit] in year i in patients, so the population served by a given hospital does month t therefore depends upon the vector of co- not fluctuate linearly with regional population growth. efficients b, which contains the effects of climate vari- g. Model validation ables collected in the covariate matrix Xit, and upon the effects of the unobserved seasonally varying factors, s(t). The robustness and accuracy of models were assessed The GAM is fitted and b and the parameters for s(t) are using a threefold cross-validation technique, the root- estimated. For the simpler GLMs, the equation is mean-square error (RMSE), and the skill score (Murphy identical except that the additive function s(t) is re- 1988). All three statistics were computed for observed moved if compared to the GAM. versus predicted values for each model. To perform the cross validation, we partitioned the data into three f. Model fitting consecutive subsets of equal length. We successively To understand the sensitivity of models to a variety excluded one of these subsets, fitted the model on the of a priori model choices, we fit three GAMs (denoted remaining data and computed the fitted values for the as models A, B, C) and a simpler GLM (denoted as excluded subset. We then combined the fitted values model D). Model A was fitted with the combination of that were obtained into one time series for ease of com- both 1-month lagged and nonlagged climatic variables parison with the ‘‘full model’’ (i.e., based on all 22 years of

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 68 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

FIG. 3. Estimated s(t) across 12 months, for the (left) models A and (right) B. Shaded area shows the 95% conﬁdence interval about s(t).

data). The skill score provides a measure of the prediction where in the numerator Xt,q is the long-term mean of X accuracy of the models by comparing the models’ pre- for a given month, t, and q denotes a particular variable dicted RMSE, Epre, with that of a reference model Eref, within the vector of X (e.g., Tmin), and bq is the co- written as efficient that corresponds to that particular variable. In the denominator s(t) is the month-specific additive Skill score 5 1 2 (E /E ). n X jb j pre ref function (which does not vary by year) and åy51 t,y y is the sum of all other model terms Xb over all variables In this case, Epre represents the RMSE of the monthly from y 5 1ton (maximum and minimum temperatures, model-predicted cases compared to the observed cases, relative humidity, dust, etc.), for a given month t. The while Eref represents the RMSE of the long-term absolute values are taken for the coefficients, b and s(t) monthly mean of the observed meningitis cases, also in the denominator, in order to omit negative terms in compared to the observed cases for each month and the equation; otherwise, the RI of a given term may be year. The reference model is thus a persistence model: if inflated due to cancellation of negative and positive one did not have a model, then they could predict cases terms. for a given year and month by assuming that they will h. Sensitivity tests equal the long-term average of cases for that month, based on observations from other years. The skill score We performed three tests of model sensitivity to 1) the is the percentage of improvement or deterioration of influence of accounting for vaccination campaigns in the a given model’s RMSE with respect to the reference case data, 2) the influence of including CO emissions as model. an explanatory variable, and 3) whether the model can To gain a perspective on the average comparative importance of each covariate for a given month, we TABLE 2. Estimates for model A. Asterisk means that variables are compute the relative influence (RI) by estimating the lagged by 1 month. effect of each covariate with respect to all the covariates in the model, based on the long-term monthly means Model A of each covariate over the 22-yr period, as well as the Predictors Coef Std error P value monthly value of s(t). The RI for a GAM is calculated as Monthly-mean Tmax (C8) 0.085 0.007 ,0.001 a percentage of all terms for a given month as follows: Monthly-mean Tmin (C8) 0.055 0.008 ,0.001 8 9 Rain (mm) 20.008 0.001 ,0.010 > > Relative humidity (%) 20.036 0.003 ,0.001 > > < X b = Dusty days (%)* 0.016 0.001 ,0.001 q q 2 5 t, Wind speed (km h 1)* 0.010 0.005 ,0.010 RI 100> n >, > > Sunshine (h) 0.047 0.015 ,0.001 :> js(t)j 1 å X jb j ;> t,y y Previous month cases 0.004 0.001 ,0.001 y51

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 69

TABLE 3. Estimates for model B. VE is vaccine efficacy, PV is the proportion of population vaccinated, and PNV is the proportion of the Model B population not vaccinated; VE is defined to be 85% in Predictors Coef Std error P value the month following vaccination administration (e.g., Monthly-mean Tmax (C8) 0.048 0.006 ,0.001 Reingold et al. 1985), and it decreases linearly to zero 8 , Monthly-mean Tmin (C ) 0.025 0.008 0.050 over the next 30 months in order to simulate for the Rain (mm) 20.003 0.001 ,0.001 Relative humidity (%) 20.015 0.003 ,0.010 development of postvaccination immunity and dimin- Dusty days (%) 0.012 0.000 ,0.001 ishing effect. The choice of 30 months was made because 2 Wind speed (km h 1) 0.010 0.004 ,0.001 polysaccharide A vaccine has an efficacy of approxi- Sunshine (h) 0.135 0.014 ,0.001 mately 2–3 years or even less in children under 4 years of , Previous month cases 0.0024 0.002 0.001 age (e.g., Reingold et al. 1985). Because of uncertainties about PV, we calculated EC for different plausible bounds of PV, 40% and 60%. be applied outside our immediate study area. The first test requires greater description here. We could not account for vaccination efficacy carried 3. Results out within the study period directly in our models be- a. Model performance cause we do not have a reliable record of the number of people vaccinated and the time of administration for any Generally, both the individual and aggregated counts of the four campaigns. However, to test the sensitivity of of the monthly hospital-reported meningitis cases ex- the model fit and accuracy to the influence of vaccina- hibit a marked annual cycle, with yearly disease onset tion, we refit the model to an adjusted case dataset that and maxima occurring during the beginning of the dry accounts for vaccinations. The proportion of cases as- season (November) and the peak of the hottest months sumed to be mitigated during epidemic years by the (March and April). Also, as in other places in the men- polysaccharide A vaccine were estimated based on in- ingitis belt, cases decrease with the increase of humidity formation collected from the FMoH indicating the years and the onset of the rainy season. See appendix A shows and districts in which vaccination campaigns were con- the individual contribution of each hospital. ducted. The month in which the vaccination campaign The shape of s(t) is shown in Fig. 3 for the best models, began was estimated to be the first month during an A and B. The shape of the estimated function of both FMoH-indicated vaccination year in which the average models is similar and follows the seasonality of the dis- case rate exceeded 20 per 100 000 persons (this was re- ease, with the months of February–April having the quired because FMoH did not indicate the month in highest values of s(t). The model estimates and standard which vaccination was begun). The threshold of 20 cases errors for models A and B are presented in Tables 2 per 100 000 in a given month was chosen because we and 3, respectively. Models C and D are not shown were limited to monthly resolved case data, whereas the for brevity, but their estimates are similar (their skill is officially defined threshold for epidemics (which trigger discussed below). Current maximum and minimum tem- reactive vaccination campaigns) is 10 cases per 100 000 peratures, sunshine duration, and 1-month lagged dust based on weekly data. Considering the strong weekly frequency and wind speed are positively correlated with variability of case counts, it is likely that at least one disease incidence, while relative humidity and precip- week in a given month would have epidemic conditions itation are inversely correlated in all models. Figure 4 (10 cases per 100 000) if 20 cases per 100 000 or more are shows the 22-yr time series of observed cases versus reported for a given month. predicted cases for models A, B, and D (results were The proportion of cases mitigated due to vaccination similar for model C). campaigns was estimated by comparing the number of The threefold cross-validation technique gives us a actual cases occurring in a given month with a projection way to verify our results (Table 4), as well as an indication of the number that may have occurred supposing poly- of how sensitive our model fits are to the period we se- saccharide vaccination was not administered, using the lected for development; for example, if the model fits following equation (Leake et al. 2002; Pinner et al. 1992): were substantially different in the early 1990s versus the late 2000s, the ‘‘full’’ versus cross-validated statistics EC 5 OC[(1 2 VE)(PV) 1 PNV], would be notably different. As Table 4 shows, the fit between the observations and both the full and the cross- where EC is the monthly estimate of expected cases if validated model statistics are remarkably good and almost vaccination was not carried out, OC is observed cases, similar, indicating that the meningitis case dynamics were

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 70 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

FIG. 4. Fit for models A, B, and D (dashed: observed, solid: predicted). Models A and D are fitted with a combination of both lagged and nonlagged climate covariates, while model B is fitted with only 1-month lagged climate covariates. Models A and B are GAMs, while D is a simpler GLM. similar throughout the 22-yr period. As expected, model suggesting it may be useful for exploring how cases may A is the best model as measured by cross-validation change in the future as a function of climate change. correlation (CVC) and skill score (0.75 and 0.52, re- Model D (the GLM) has a lower CVC and skill score spectively). The skill score indicates that the RMSE of (0.62 and 0.40, respectively) compared to the GAMs, model A is 52% lower than if one assumes the number of suggesting that the GAM additive function s(t) is effective cases in a given year and month are equal to the long- in capturing some of the unobserved seasonally varying term monthly average of observed cases for that month; factors that may affect meningitis. thus, model A exhibits substantial skill compared to The relative influence of each explanatory variable for doing no modeling at all. The 1-month lag model B also the four months in which meningitis incidence is gen- has similar CVC and skill score statistics (0.73 and 0.45, erally high is shown in Fig. 5. Overall, the maximum respectively), suggesting its potential for short-term pre- temperature shows a comparatively important influence diction of cases. Likewise, model C has similar statistics, on the fitted meningitis cases for models A and B.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 71

TABLE 4. Model validation results, given by correlation and skill remains almost the same across the months, having an score. Full and CV stand for full and threefold cross-validated important negative influence of about 215%; this is models, respectively. because humidity usually remains consistently low Kendall correlation Skill score throughout the dry season. As expected, the dust and Models Full CV Full CV wind speed influence is higher in the cold dry months in which the Harmattan is strongest. Rainfall plays A 0.762 0.746 0.527 0.515 B 0.740 0.732 0.458 0.449 a negligible role throughout these months because it C 0.750 0.736 0.489 0.476 rarely rains (Fig. 1). The number of cases in the pre- D 0.634 0.616 0.419 0.401 vious month provides a comparatively modest influence of up to about 10%. The function s(t), which accounts for unobserved seasonally varying explana- Regardless of the model, the relative influence of max- tory variables, varies in influence from about 6% to imum temperature increases from about 30% in January 15%, suggesting that unexplained factors are impor- to over 35% in April during the peak of meningitis in- tant drivers of meningitis incidence, especially during cidence. The influence of average relative humidity the annual peak of cases.

FIG. 5. Relative inﬂuence of explanatory variables on meningitis in northwest Nigeria for models A and B, for the dry-season months in which the meningitis burden is typically highest: January, February, March, and April.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 72 WEATHER, CLIMATE, AND SOCIETY VOLUME 6 b. Model sensitivity

1) EXPERIMENT 1 As described in section 2h, to investigate the influence on our results of the four vaccination campaigns that occurred between 1990 and 2011, we estimated the expected cases that would have occurred in the absence of the campaigns. A model was fit using the expected cases as predictands and retaining all of the explanatory variables used in model A, except the previous month’s cases. We omitted previous cases because we intend to employ this model in a follow-up study to investigate the impacts of FIG. A1. African meningitis belt (WHO Working Group 1998). climate change on meningitis in the absence of vaccination (because it is impossible to know what vaccine ad- for Kaduna, and then compare the predicted time series vancements will occur in the future). The new model is with that observed from Kaduna hospital. Although the therefore identical to model C, except it is trained on the fit is not as good as that of model A (CVC 5 0.57 and expected case data rather than the actual case data. The skill score 5 0.39), the monthly variability of cases is 5 new model (with PV 40%) has a higher CVC and skill reasonably captured. We also fitted a completely new score of 0.75 and 0.52, respectively, compared to model C GAM with four DOF specifically for Kaduna city, em- (0.74 and 0.48, respectively). The slightly higher skill score ploying the same variables as for model A. The esti- of the new model compared to model C is perhaps ex- mates of this model are shown in appendix C. The only pected, since the additional effects of vaccination are at difference of this model with model A is that the rain least partially removed, and therefore the explanatory variable is less significant, but the model remains qual- climate variables are targeting a largely expected number itatively similar in terms of predictive power. This sug- of meningitis cases without vaccination. Since the choice gests that the explanatory variables employed in our of PV is somewhat arbitrary, we also fitted a model as- models may be relevant for simulating meningitis in- suming a different proportion of the population was vac- cidence in other regions of Nigeria, and perhaps the 5 cinated, PV 60%, and found that the values and broader Sahelian region [this is supported by the fact significance of the model coefficients in the GAM are that Dukic et al. (2012) found similar explanatory vari- relatively insensitive to the choice of PV (see appendix B), ables when modeling meningitis incidence in northern 5 although the best model fit is for the case of PV 40%. Ghana]. 2) EXPERIMENT 2 Air quality contributes to the risk and incidence of 4. Discussion meningitis in northern Ghana (e.g., Hodgson et al. 2001). We investigate this in northwest Nigeria by refitting In this study, we employed both GAMs and GLMs to model A, keeping all of the variables in the model but model the influence of meteorological conditions on the now including the total regional CO from biomass monthly meningitis incidence from three hospitals in burning emission estimates (for the 15-yr period from northwest Nigeria, using monthly aggregate counts of 1997 to 2009, for which we have CO data). Although the clinically diagnosed hospital-reported meningitis cases previous month’s CO emissions are significant in the for 1990–2011. The case data exhibit a marked annual model, it does not substantially change the model esti- cycle, with yearly disease maxima occurring during the mates for other explanatory variables, and it does not peak of the hot dry season in March and April. Ex- appreciably change the model fit (see appendix B). planatory variables in the models include meteorologi- Therefore, biomass burning CO emissions, and presumably cal variables, cases from the previous month, and s(t)in other air pollutants released in conjunction with the CO, the GAMs. Model performance was estimated by three- do not appear to exert much influence on meningitis in- fold cross validation, RMSE, and skill score. cidence in northwest Nigeria compared to the other The results indicated that both explanatory (fitted meteorological variables in the model. with the combination of lagged and nonlagged climatic variables) and predictive (fitted with only 1-month lagged 3) EXPERIMENT 3 climatic variables) models showed similar capabilities to For the purpose of testing the model elsewhere, we fit values of meningitis incidence. The best explanatory use the estimated coefficient of model A to predict cases model (A) had a CVC of 0.75 with the observations and

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 73

TABLE A1. Experiment 1, showing model estimates for two bounds of PV. Asterisk means that variables lagged by 1 month.

PV 5 40% PV 5 60% Predictors Coef Std error P value Coef Std error P value Monthly-mean Tmax (C8) 0.0827 0.0069 ,0.001 0.0821 0.0067 ,0.001 Monthly-mean Tmin (C8) 0.0576 0.0078 ,0.001 0.0565 0.0076 ,0.001 Rain (mm) 20.0039 0.0006 ,0.001 20.0035 0.0005 ,0.001 Relative humidity (%) 20.0403 0.0024 ,0.001 20.0431 0.0023 ,0.001 2 Wind speed (km h 1)* 0.0143 0.0048 ,0.001 0.0165 0.0047 ,0.001 Sunshine (h) 0.0494 0.0124 ,0.001 0.0476 0.0120 ,0.001 Dusty days (%)* 0.0166 0.0007 ,0.001 0.0176 0.0007 ,0.001 a skill score of 0.52 out of 1.0. Additionally, the cross- improve for future studies, as Nigerian authorities are validation version of model A (that was developed by now keeping detailed records of vaccination campaigns systematically omitting the first, middle, and last 7 years for the ongoing MenAfriVac (the new conjugate A of case data) exhibited nearly identical statistics to the vaccine). The recent change in strategy from reactive to full model that was fit using all 22 years of case data, preventive vaccination is a welcome development in the indicating that the model performance is not sensitive to effort of controlling the disease; however, there is need the period chosen for development. Although the best for continual surveillance because other strains not cov- predictive model B was not as powerful as the best ex- ered by the conjugate vaccine may cause epidemics (e.g., planatory model A in terms of fitting performance, it has Koumare et al. 1993). strong potential for short-term disease prediction, hav- Although, there is no agreed-upon physical explana- ing a CVC of 0.73 and a skill score of 0.45. It is note- tion for the role of meteorological conditions in the in- worthy that all of the most successful models were cidence of meningitis (e.g., Cheesbrough et al. 1995), our GAMs, suggesting that model fit is substantially im- results support the hypothesis that hot, dry, dusty con- proved by employing the monthly varying s(t). Despite ditions may facilitate both the transmission and the de- the fact that reliable vaccination data are difficult to velopment of invasive meningitis in northwest Nigeria. obtain, we estimated the expected number of cases that For example, these conditions might be playing impor- would have occurred assuming the vaccination cam- tant role during the cold dry months, aiding in initiating paigns were not carried out. We then fit a model to these the meningitis season by causing microtrauma to the nasal expected cases and found that the model fit and accuracy mucosa (Burgess and Whitelaw 1988). This damage may changed very little, suggesting the results are reasonably make it possible for the bacteria to penetrate the naso- insensitive to whether vaccination is accounted for. Re- pharyngeal membrane and subsequently enter the blood gardless, documentation of vaccination campaigns may stream, causing invasive disease. This may explain why

FIG. B1. Reported cases of meningitis from three hospitals in Kano, Sokoto, and Gusau (dashed) in comparison with reported cases at the WHO district level from corresponding districts (solid) between 2007 and 2011.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 74 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

TABLE B1. Experiment 2, showing estimates for the model with variability in northwest Nigeria, and also emphasize the CO emissions. Asterisk means that variables lagged by 1 month. importance of additional risk factors that are not well Predictors Coef Std error P value understood but may be linked to societal and behavioral practices. With respect to the latter, we are currently Monthly-mean Tmax (C8) 0.0841 0.0072 ,0.001 Monthly-mean Tmin (C8) 0.0532 0.0018 ,0.001 limited only to collectively accounting for these un- Rain (mm) 20.0103 0.0021 ,0.010 observed seasonally varying climatic and nonclimatic Relative humidity (%) 20.0415 0.0032 ,0.001 risk factors via functions such as s(t). Identifying and Dusty days (%)* 0.0163 0.0014 ,0.001 quantifying such factors and improving disease surveil- 21 , Wind speed (km h )* 0.0106 0.0005 0.010 lance will likely enhance our understanding and ability Sunshine (h) 0.0334 0.0105 ,0.001 Previous month cases 0.0065 0.0042 ,0.001 to predict and reduce meningitis incidence. CO emission* 0.0096 0.0122 ,0.010 Acknowledgments. We thank the World Health Or- reported meningitis cases are highest during the hot dry ganization, the Nigerian National Primary Health Care period (February–May) that follows the cold dry period. Development Agency, the Nigerian Ministry of Health, Unobserved seasonally varying nonclimatic factors the Nigerian Meteorological Agency, and the manage- such as the occurrence of URTIs, and societal and be- ment of the four state-owned hospitals for providing havioral practices, which are represented in a very basic data and assistance. Also, we thank the editor and the four sense by the function s(t), are likely to enhance the anonymous reviewers, whose useful comments helped transmission of the disease in this region. For example, significantly in improving the article. Christine Wiedinmyer during the cold dry period (which favors diseases such as (NCAR) provided emissions data. Funding for this work influenza) people are often overcrowded in rooms and came from the Nigerian Tertiary Education Trust Fund, sometimes cluster around wood fires for warming. This Google.org, and the U.S. National Science Foundation overcrowding (e.g., Brundage and Zollinger 1987) might (Grant GEO 1211668). This study was part of the first enhance transmission through respiratory droplets, author’s doctoral dissertation, to be submitted to the Uni- while particulates from wood fires may irritate the lining versity of Birmingham, United Kingdom. Finally, we de- of the nasal mucosa. Additionally, social gatherings such clare no actual or potential competing financial interest. as marriages and economic activities in market places tend to be more frequent and active during this season. APPENDIX A

5. Conclusions Meningitis Belt Map and Vaccination Inﬂuence Our results indicate the role of speciﬁc meteorological A map of the African meningitis belt is shown in conditions in explaining and predicting monthly meningitis Fig. A1 Experiment 1 is detailed in Table A1.

FIG. C1. Distribution of meningitis cases from the three selected hospitals in northwest Nigeria.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC JANUARY 2014 A B D U S S A L A M E T A L . 75

TABLE C1. Experiment 3, showing model estimates for Kaduna. Cookson, S., J. Corrales, J. Lotero, M. Regueira, N. Binsztein, Asterisk means that variables are lagged by 1 month. M. Reevas, G. Ajello, and W. R. Jarvis, 1998: Disco fever: Epidemic meningococcal disease in northeastern Argentina Predictors Coef Std error P value associated with disco patronage. J. Infect. Dis., 178, 266–269. Monthly-mean Tmax (C8) 0.0872 0.0042 ,0.001 Cornforth, R., 2012: Overview of the West African monsoon 2011. Monthly-mean Tmin (C8) 0.0613 0.0023 ,0.001 Weather, 67, 59–65. Rain (mm) 20.0054 0.0031 ,0.050 Decosas, J., and J. B. T. Koama, 2002: Chronicles of an outbreak Relative humidity (%) 20.0462 0.0034 ,0.001 foretold: Meningococcal meningitis W135 in Burkina Faso. Dusty days (%)* 0.0285 0.0016 ,0.001 Lancet Infect. Dis., 2, 763–765. 2 Wind speed (km h 1)* 0.0166 0.0040 ,0.010 Dukic, V., and Coauthors, 2012: The role of weather in meningitis Sunshine (h) 0.0327 0.0150 ,0.001 outbreaks in Navrongo, Ghana: A generalized additive mod- Previous month cases 0.0116 0.0123 ,0.001 eling approach. J. Agric. Biol. Environ. Stat., 17, 422–460, doi:10.1007/s13253-012-0095-9. Fischer, M., and Coauthors, 1997: Tobacco smoke as a risk factor for menigococcal disease. Pediatr. Infect. Dis. J., 16, 979–983. APPENDIX B Greenwood, B. M., 2006: Editorial: 100 years of epidemic meningitis in West Africa—Has anything changed? Trop. Med. Int. Meningitis Cases in Northwest Nigeria and the Role Health, 11, 773–780. ——, and J. M. Stuart, 2012: A vaccine to prevent epidemic men- of Biomass Burning ingitis in Africa. Lancet Infect. Dis., 12, 738–739. The cases of meningitis reported compared with WHO ——, A. K. Bradley, I. S. Blakebrough, S. Wali, and H. C. Whittle, 1984: Meningococcal disease and season in sub-Saharan Af- district level cases from the period 2007–11 are shown in rica. Lancet, 323, 1339–1342. Fig. B1. Experiments 2 described in Table B1. Harrison, L. H., C. L. Trotter, and M. E. Ramsey, 2009: Global epidemiology of meningococcal disease. Vaccine, 27, 51–63. Hastie, T. J., and R. J. Tibshirani, 1999: Generalized Additive APPENDIX C Models. Chapman and Hall Ltd., 335 pp. Hodgson, A., T. Smith, S. Gagneux, M. Adjuik, G. Pluschke, Individual Hospital Meningitis Counts and Kaduna N. K. Mensah, and B. Genton, 2001: Risk factors for meningococcal meningitis in northern Ghana. Trans. Roy. Soc. Trop. Model Med. Hyg., 95, 477–480. The individual counts of meningitis reported from Jandarov, R., M. Haran, and M. Ferrari, 2012: A compartmental model for meningitis: Separating transmission from climate Kano, Sokoto, and Gusau are displayed in Fig. C1. Ex- effects on disease incidence. J. Agric. Biol. Environ. Stat., 17, periment 3 is described in Table C1. 395–416. Koumare, B., F. Bougoudogo, M. Cisse, T. Doumbia, and M. M. Keita, 1993: Bacteriological aspects of purulent men- REFERENCES ingitis in Bamako district. A propos of 1541 bacterial strains Besancenot, J. P., M. Boko, and P. C. Oke, 1997: Weather condi- collected from 1979 to 1991. Bull.Soc.Pathol.Exot.,86, 136–140. tions and cerebrospinal meningitis in Benin (Gulf of Guinea, Kristiansen, P. A., and Coauthors, 2011: Baseline meningococcal West Africa). Eur. J. Epidemiol., 13, 807–815. carriage in Burkina Faso before the introduction of a menin- Broome, C. V., M. A. Rugh, A. A. Yada, L. Giat, H. Giat, gococcal serogroup A conjugate vaccine. Clin. Vaccine Im- J. M. Zeltner, W. R. Sanborn, and D. W. Fraser, 1983: Epi- munol., 18, 435–443. demic group C meningococcal meningitis in Upper Volta. Lapeyssonnie, L., 1963: Cerebrospinal meningitis in Africa. Bull. Bull. World Health Organ., 61, 325–330. World Health Organ., 28 (Suppl.), 1–114. Brundage, J., and W. Zollinger, 1987: Evolution of meningococcal Leake, J. A., and Coauthors, 2002: Early detection and response to disease epidemiology in the U.S. Army. Evolution of Menin- meningococcal disease epidemics in sub-Saharan Africa: Ap- gococcal Disease, N. A. Vedros, Ed., Vol. 1, CRC Press, 5–25. praisal of the WHO strategy. Bull. World Health Organ., 80, Burgess, K. R., and W. A. Whitelaw, 1988: Effects of nasal cold 342–349. receptors on pattern of breathing. J. Appl. Physiol., 64, 371–376. Lingappa, J. R., and Coauthors, 2003: Serogroup W-135 meningo- Cameron, A. C., and P. K. Trivedi, 1998: Regression Analysis of coccal disease during Hajj, 2000. Emerg. Infect. Dis., 9, 665–671. Count Data. Econometric Society Monogr., No. 30, Cambridge Maiden, M. C., and Coauthors, 2008: Impact of menigococcal University Press, 436 pp. serogroup C conjugate vaccines on carriage and herd im- Cartwright, K. A. V., 1995: Meningococcal carriage and disease. munity. J. Infect. Dis., 197, 737–743. Meningococcal Disease, K. A. V. Cartwright, Ed., John Wiley Mohammed, I., A. Nasidi, A. S. Alkali, M. A. Garbati, E. K. Ajayi- and Sons, 115–146. Obe, K. A. Audu, A. Usman, and S . Abdullahi, 2000: A severe Cheesbrough, J. S., A. P. Morse, and S. D. R. Green, 1995: Me- epidemic of meningococcal meningitis in Nigeria. Trans. Roy. ningococcal meningitis and carriage in western Zaire: A hy- Soc. Trop. Med. Hyg., 94, 265–270. poendemic zone related to climate. Epidemiol. Infect., 114, Molesworth, A. M., M. C. Thomson, S. J. Connor, M. P. Cresswell, 75–92. A. P. Morse, P. Shears, C. A. Hart, and L. E. Cuevas, 2002: Climate Research Division, cited 2008: RClimDex 1.0. Expert Where is the meningitis belt? Deﬁning an area at risk of epi- team on climate change detection and indices. [Available demic meningitis in Africa. Trans. Roy. Soc. Trop. Med. Hyg., online at http://etccdi.paciﬁcclimate.org/software.shtml.] 96, 242–249.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC 76 WEATHER, CLIMATE, AND SOCIETY VOLUME 6

Moore, P. S., J. Hierholzer, W. Dewitt, K. Gouan, D. Djore, and Thomson, M. C., A. M. Molesworth, M. H. Djingarey, K. R. Yameogo, T. Lippeveld, 1990: Respiratory viruses and mycoplasma as F. Belanger, and L. E. Cuevas, 2006: Potential of environmental cofactors for epidemic group A meningococcal meningitis. models to predict meningitis epidemics in Africa. Trop. Med. Int. JAMA, J. Amer. Med. Assoc., 264, 1271–1275. Health, 11, 781–788. Mueller, J. E., S. Yaro, Y. Madec, P. K. Somda, R. S. Idohou, and Trotter, C. L., and B. M. Greenwood, 2007: Meningococcal car- B. M. Njanpop Lafourcade, 2008: Association of respiratory riage in the African meningitis belt. Lancet Infect. Dis., 7, 797– tract infection symptoms and air humidity with meningococcal 803. carriage in Burkina Faso. Trop. Med. Int. Health, 13, 1543–1552. van der Werf, G. R., and Coauthors, 2010: Global fire emissions Murphy, A. H., 1988: Skill scores based on the mean square error and the contribution of deforestation, savanna, forest, agri- and their relationships to the correlation coefficient. Mon. cultural, and peat fires (1997–2009). Atmos. Chem. Phys., 10, Wea. Rev., 116, 2417–2424. 11 707–11 735. Palmgren, H., 2009: Meningococcal disease and climate. Global Vestrheim, D. F., E. A. Høiby, I. S. Aaberge, and D. A. Caugant, Health Action, 2, doi:10.3402/gha.v2i0.2061. 2010: Impact of a pneumococcal conjugate vaccination pro- Peltola, H., 1983: Meningococcal disease—Still with us. Rev. Infect. gram on carriage among children in Norway. Clin. Vaccine Dis., 5, 71–91. Immunol., 17, 325–334. Pinner, R. W., and Coauthors, 1992: Epidemic meningococcal Wahdan, M. H., and Coauthors, 1973: A controlled field trial of disease in Nairobi, Kenya, 1989. The Kenya/Centers for Dis- a serogroup A meningococcal polysaccharide vaccine. Bull. ease Control (CDC) meningitis study group. J. Infect. Dis., World Health Organ., 48, 667–673. 166, 359–364. WHO, 2010: Managing meningitis epidemic in Africa: A quick R Core Team, 2009: R: A Language and Environment for Statistical reference guide for health authorities and health-care Computing. R Foundation for Statistical Computing, 409 pp. workers. WHO/HSE/GAR/ERI/2010.4, 23 pp. [Available Reingold, A. L., and Coauthors, 1985: Age-specific differences in online at http://whqlibdoc.who.int/hq/2010/WHO_HSE_ duration of clinical protection after vaccination with menin- GAR_ERI_2010.4_eng.pdf.] gococcal polysaccharide a vaccine. Lancet, 2, 114–118. ——, cited 2012: Meningococcal meningitis. [Available online at Riou, J.-Y., S. Djibo, L. Sangare, J.-P. Lombart, P. Fagot, http://www.who.int/mediacentre/factsheets/fs141/en/.] J.-P. Chippaux, and M. Guibourdenche, 1996: A predictable WHO Working Group, 1998: Control of epidemic meningococcal comeback: The second epidemic of infections caused by disease: WHO practical guidelines. 2nd ed. WHO/EMC/BAC/ Neisseria meningitidis serogroup A subgroup III in Africa, 98.3, 84 pp. 1995. Bull. World Health Organ., 74, 181–187. World Bank, cited 2012: Data: Nigeria. [Available online at http:// Rosenstein, N. E., B. A. Perkins, D. S. Stephens, T. Popovic, and data.worldbank.org/country/nigeria.] J. M. Hughes, 2001: Meningococcal disease. N. Engl. J. Med., World Meteorological Organization, 1995: Manual on codes. 344, 1378–1388. Vol. I.1, Part A: Alphanumeric Codes, WMO Publ. 306, 503 pp. Schuchat, A., K. Robinson, J. D. Wenger, L. H. Harrison, [Available online at http://www.wmo.int/pages/prog/www/ M. Farley, A. L. Reingold, L. Lefkowitz, and B. A. Perkins, WMOCodes.html.] 1997: Bacterial meningitis in the United States in 1995. Yaka, P., B. Sultan, H. Broutin, S. Janicot, S. Philippon, and N. Engl. J. Med., 337, 970–976. N. Fourquet, 2008: Relationships between climate and year- Sultan, B., 2005: Influence of climate upon the meningitis onset in to-year variability in meningitis outbreaks: A case study in West Africa. Med. Sci., 21, 470–471, doi:10.1051/medsci/ Burkina Faso and Niger. Int. J. Health Geogr., 7, doi:10.1186/ 2005215470. 1476-072x-7-34.

Unauthenticated | Downloaded 10/01/21 11:58 PM UTC