International Journal of Epidemiology© International Epidemiological Association 1992

Vol. 21, No. 1Printed in Groat Britain

Detection of Meningitis Epidemics inAfrica: A Population-Based AnalysisPATRICK S MOORE.* BRIAN D PLIKAYTIS,* GAIL A BOLAN,* MARGARET J OXTOBY,#

ADAMOU YADA,»# ALAIN ZOUBGA." ARTHUR L REINGOLD* AND CLAIRE V BROOME*

Moore P S (Meningitis and Special Pathogens Branch, Center for Infectious Diseases, Centers for Disease Control,Atlanta, GA, 30333 USA), Plikaytis B D, Bolan G A, Oxtoby M J, Yada A, Zoubga A, Reingold A L and Broome C V.Detection of meningitis epidemics in Africa: a population-based analysis. International Journal of Epidemiology 1992;21: 155-162.Portions of sub-Saharan Africa are subject to major epidemics of meningococcal meningitis that require early detectionand rapid control. We evaluated the usefulness of weekly meningitis rates derived from active surveillance data inBurkina Faso for detecting a meningitis epidemic. By analysing the rates of disease in 40 x 40km2 areas within a studyregion of Burkina Faso, we found that a threshold of 15 cases/100 000/week averaged over 2 weeks was 72-93% sen-sitive and 92-100% specific in detecting epidemics exceeding 100 cases/100000/year. During epidemic periods, thepositive predictive value of this threshold approached 100% for detecting local epidemics. Additionally, meningitisincidence was proportional to village size, with villages greater than 8000 having the highest disease rates during amajor group A meningococcal epidemic in 1983-1984. Despite the rudimentary nature of surveillance data available inmany developing countries, these data can be used to detect the early emergence of meningitis epidemics. Additionalstudies are needed to determine the relevance of this approach for detecting epidemics.

Epidemic group A meningococcal meningitis is a majorhealth problem in sub-Saharan Africa. Since thepolysaccharide meningococcal vaccine has a limitedduration of protection in young children,1 routinemeningococcal vaccination is not effective in prevent-ing epidemics. However, widespread vaccination cancontrol meningococcal disease if implemented early inthe course of an epidemic.2'3 Epidemic control requiresthat an epidemic be detected early and that vaccinationefforts be organized rapidly.4 Because vaccinationcampaigns are expensive and divert resources fromongoing health programmes, mounting a campaign inthe absence of an epidemic can have an adverse impacton health-care delivery in developing countries. A

'Meningitis and Special Pathogens Branch, Center for InfectiousDiseases, US Public Health Service, Centers for Disease Control,Atlanta, GA 30333, USA.••Ministry of Public Health, Ouagadougou, Burkina Faso.Current addresses: PSM: Arboviral Diseases Branch, Centers forDisease Control, Fort Collins, CO, USA; GAB: San Francisco Depart-ment of Health, San Francisco, CA, USA; MJO: Division ofHIV/AIDS, Centers for Disease Control, Atlanta, GA, USA: AY andAZ: Ministre de la Sante Publique, BP 1046, Ouagadougou, BurkinaFiio; ALR: Department of Epidemiology and Biostatistks, Universityof California, Berkeley, CA, USA; Reprint requests to: Dr C VBroome.NB. Use of trade names is for identification only and does not implyendorsement by the US Public Health Service or by the US Depart-ment of Health and Human Services.

simple and reliable method of detecting emergingepidemics in developing countries is clearly needed.

Although epidemics occur in 8-14 year cycles in theAfrican 'meningitis belt', predicting the exact timingand location of an epidemic is not currently possible.Several methods have been used to predict the occur-rence of epidemics based on disease patterns fromprevious years.'•' These methods may be useful fordetermining the likelihood of an epidemic during agiven year, but further documentation of their predic-tive value is needed. In the meantime, early detectionof an emerging epidemic based on meningitis sur-veillance data is a reasonable objective.

In most nations of the African meningitis belt, men-ingitis is one of the diseases included in the emergencynotification system and is reported to public healthagencies. These surveillance data, however, are oftenrudimentary and have not been rigorously evaluated asto whether they are a useful measure of disease activity.To detect an emerging epidemic, surveillance must besensitive to changes in meningococcal meningitis in-cidence. Since laboratory facilities are uncommon inareas that are severely affected by epidemics, men-ingitis cases are generally diagnosed on clinicalgrounds alone without laboratory confirmation of in-fection. This clinical case definition does not differen-tiate group A meningococcal meningitis from othercauses of sporadic meningitis. While meningococcal

155

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

156 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

meningitis epidemics are easy to identify after the fact,it is unclear whether surveillance can reliably detect theearly emergence of epidemics.

One method for predicting the emergence of an epi-demic is to determine whether or not weekly meningitisincidence thresholds exist that are generally exceededduring epidemics, but not during endemic periods.Developing a weekly or biweekly threshold rate thataccurately predicts the ultimate occurrence of anepidemic may allow the early mobilization of resourcesto combat the outbreak. Currently, no criteria exist fordetermining when to initiate vaccination campaignsbased on ongoing surveillance data. We conducted anactive review of population-based meningitis sur-veillance data from 1979 to 1984 in Burkina Faso todetermine the utility of using such data for the earlydetection of epidemics. Using weekly and biweeklydisease rates from surveillance data, we developedthreshold meningitis-rate guidelines that were bothsensitive and specific for detecting the eventualemergence of an epidemic. In addition, we used thesedata to investigate the relationship between village sizeand meningitis incidence in Burkina Faso.

METHODSMeningitis SurveillanceWe reviewed records at local health-care facilities andpublic health centres for clinically defined meningitiscases in a central region of Burkina Faso from July1979 to June 1984. Meningitis surveillance recordswere collected from community health centres, dispen-saries and hospitals in the four central provinces(Sissili, Sanguie, Bulkiemde, and Passore) making upthe study region (Figure 1). This region had a popula-

Burklna Faso

C O T E D I V O I R E

FIGURE 1 Study area in central Burkina Faso.

tion of 890000 (mid-1981 estimate) distributed over656 known villages and towns.

For the purposes of this study, all patients seen attreatment centres with a diagnosis of meningitis wereconsidered to be cases and the date of presentation tothe centre was considered to be the date of onset.Because most of the treatment centres did not havelaboratory facilities, meningitis cases were primarilydiagnosed by the characteristic signs and symptomsof meningitis (fever, headache, nuchal rigidity andpetechial rash). This diagnosis was often confirmed byvisual examination of cerebrospinal fluid for turbidity.Although meningitis cases included in our study prob-ably represent a mixture of meningococcal, Haemophilus,pneumococcal, and nonbacterial meningitides, menin-gococcal meningitis is the only cause of epidemic men-ingitis in this region.

Meningitis Incidence CalculationsSince the meningitis season in Burkina Faso extendsfrom December to May, annual incidence rates werecalculated from 1 July to the following 30 June inorder to include the entire meningitis season within 1year. Village populations in the study region weredetermined from a 1975 census and adjusted for a2.3% annual growth rate (World Bank population esti-mates, 1983). Incidence calculations were made usingthe SAS software program (Statistical Analysis Sys-tems, Cary, NC) by dividing the number of casesreported in a given village by the annually-adjustedpopulation of the village. Village and health centrelocations were determined by digitizing census mapsprovided by the Ministry of Public Health (MoPH)using a HP 9845 C desktop computer (Hewlett-PackardCo., Sunnyvale, CA). Because the location, popula-tion, and number of meningitis cases for each villagewas known, disease rates in a given area could becalculated. For example, the entire study region couldbe divided into thirty-two 40 x 40km2 grids and themeningitis rate determined for each 1600km2 areausing the overall meningitis rate for the villages withineach area grid.

Meningitis Incidence and Village SizeTo determine the relationship between village size andannual meningitis rate, all 656 villages in the studyregion were divided into three broad population cate-gories (<2000, 2000-8000, and >8000 people).Average incidence rates weighted by village populationsize were determined for each of the three populationcategories for each year. To assess the influence thataccess to medical care had on surveillance data, men-ingitis rates also were examined in a subset of 37villages with active medical dispensaries or hospitals.

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

DETECTION OF MENINGITIS EPIDEMICS 157

Determination of Weekly Threshold RatesTo detect impending epidemics within the study area,the region was divided into 40 x 40km2 areas. Of the 32areas in the study region, nine with initial populationsless than 10000 were excluded from the analysis toavoid fluctuations due to small populations. Weeklymeningitis rates were examined throughout the 5-yearperiod in the remaining 23 areas and compared withannual meningitis rates for each of the 40 x 40km2

areas.To perform these calculations, an annual rate in a 40

x 40km2 area of 100 cases/100000 population orgreater was arbitrarily defined as an epidemic—a ratethat is comparable to the epidemic definition used byother authors.3 Areas with annual rates exceeding 100cases/100 000 in any given year were retrospectivelyconsidered to have experienced an epidemic, and areaswith annual rates below this cutoff were considered tohave been nonepidemic. Weekly rates between 5 and30 cases/100 000 population were then investigatedas potential thresholds for predicting the eventualemergence of an epidemic during that same year. Eacharea exceeding a given threshold weekly rate at anytime during the year was examined to see if the samearea subsequently exceeded the 'epidemic' rate of 100cases/100 000/year. By determining whether or noteach grid exceeded both the given weekly thresholdrate and the epidemic annual rate, the sensitivity, thespecificity, and the positive predictive value of aweekly threshold rate to predict epidemic annual ratescould be calculated. Threshold rates averaged over 2consecutive weeks were also calculated by dividing thebiweekly rate by 2 to determine the average weeklymeningitis rate. This analysis was performed in anattempt to improve the accuracy of the detectionmethod, since it requires that a given threshold rate bemaintained on average for at least 2 weeks in order tosurpass the threshold.

Because a major group A meningococcal meningitisepidemic occurred in Burkina Faso during 1984, the5-year observation period was divided into separateendemic and epidemic periods for the purposes of thisstudy. The endemic period consisted of the 4 yearsfrom 1 July 1979 to 30 June 1983, and the epidemicperiod was the single year from 1 July 1983 to 30 June1984. The endemic period had 92 grid-year observa-tions (23 grids over 4 years) and the epidemic periodhad 23 grid-year observations (23 grids over 1 year).

VaccinationTo determine the effect of meningococcal vaccinationon meningitis rates, vaccine distribution records wereobtained from the MoPH. No other sources are known

to have distributed meningococcal vaccine during thestudy period, although limited amounts of vaccinewere also commercially available. The vaccination ratefor 40 x 40km2 areas was determined because it was notpossible to determine the vaccination rate within indi-vidual villages. Because meningococcal vaccine efficacyremains high for several years in children over 4 yearsold and adults, cumulative vaccination rates were com-pared with meningitis rates for each year for each 40 x40km2 area. It was assumed that all doses were admin-istered within the area containing the vaccine distribu-tion centre and that no one was vaccinated twice.These assumptions tend to overestimate the amount ofvaccine administered within each area. The influenceof pneumococcal and Haemophilus vaccination wasconsidered to be negligible since these vaccines havenot been widely distributed in Burkina Faso.

RESULTSFrom July 1979 to June 1984, 5591 cases of meningitisdiagnosed on clinical grounds were recorded in healthcentres throughout the study region. Of these cases,the home village of 5478 (98%) of the patients wasrecorded. The meningitis season in a given year typi-cally lasted from 17 to 33 weeks during the studyinterval (Figure 2). During the 4-year period from July1979 through June 1983, annual endemic meningitisrates ranged from a low of 30.2 cases/100 000 in1979-1980 to a high of 94 cases/100000 in 1980-1981.

60-

°. 50-

8

i 40H

| 20H

o^79 80 84 8581 82 83

Year

FIGURE 2 Weekly meningitis rates in the study area.

In 1983-1984, a major group A meningococcal men-ingitis epidemic with 2965 cases occurred in the studyregion, resulting in an annual incidence rate of 318cases/100000. During this epidemic, high rates werenot uniformly distributed throughout the study region.When the study region was divided into thirty two 40 x40km2 areas, 17 (53%) of these areas had incidencerates greater than 100 cases/lOOOOO/year, and 12

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

158 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

(38%) of these areas had rates exceeding 200cases/100000/year. In contrast, 15 (47%) areas, pri-marily located in the sparsely populated province ofSissili, did not exceed the epidemic threshold of 100cases/100000/year during the 1983-1984 epidemic.During the endemic period 1979-1983, local outbreaksoften exceeded the 100 cases/100 000/year threshold,but were not widespread enough to increase the levelof disease throughout the study region to epidemicproportions.

Village Size and Meningitis RateThe annual meningitis rate was strongly associatedwith village population size during both endemic andepidemic years. The meningitis rate for villages withpopulations over 8000 was two- to four-fold higher thanin villages with populations under 2000 during theendemic years 1979-1983 (Figure 3a). This trend wasmarked during the 1983-1984 epidemic; the annual

a 1000

1979-80 1980-81 1981-82 1982-83 1983-84

Year

b 900

8 eoo .

IBE

400

200

1979-80 1983-84

FIGURE 3a Annual meningitis rales in villages with <2000 popula-tion, 2000SOOO population and >8000 population, b. Annual men-ingitis rates in the 37 villages in the region with health centres.

incidence rate in villages over 8000 in population was888 cases/100 000 compared with 219 cases/100000 forvillages with under 2000 population. During this year,the largest two cities in the study area had populationsgreater than 15000 and a combined annual meningitisincidence rate of 10% cases/100000, or approximately1.1% of the citywide population for 1983-1984.Although only 13.9% of the population in the fourprovinces lived in villages with 8000 population orgreater, 38.7% of 1983-1984 meningitis cases werereported from these villages.

The relationship between meningitis rate and villagesize could result from patients in larger villages havingbetter access to health care and being more likely to beseen at a health facility. However, an examination ofonly those villages with health facilities suggests thatthis was not due to a reporting artifact. When only the37 villages with health-care facilities were examined,meningitis rates remained highest in large villagesthroughout the study period (Figure 3b). During theepidemic year 1983-1984, annual meningitis rates were662 cases/100000 in villages over 8000 inhabitants withhealth facilities compared to only 55 cases/100000 invillages under 2000 inhabitants with health facilities.

Use of Weekly Rates to Detect EpidemicsTo examine the utility of predicting epidemics fromweekly threshold rates, the study region was dividedinto thirty-two 40 x 40km2 areas. Nine areas withpopulations less than 10000 were excluded from theanalysis (to avoid fluctuations due to small numbers),leaving twenty-three 40 x 40km2 areas for analysis.Threshold rates for the endemic years 1979-1983 andthe epidemic year 1983-1984 wwe examined separately.

Overall, the sensitivity of any threshold rate fordetecting local epidemics was high for epidemic years(Figure 4a), but declined for endemic years as the strin-gency of the threshold increased. In contrast, higherthreshold values increased the specificity of epidemicdetection. This trend is seen in Figure 4b for both epi-demic and endemic periods; specificity progressivelyincreases as the threshold increases from 5 to 30cases/100 000/week.

The positive predictive value (i.e. the per cent ofareas exceeding the threshold rate that actually had anepidemic annual rate) also increased with higherthreshold values (Figure 4c). As anticipated, thepositive predictive value was also higher at any giventhreshold value in 1983-1984 than in 1979-1983because the high prevalence of local outbreaks duringthe major meningococcal epidemic in 1983-1984.

Overall, a threshold rate of 15 cases/100000/weekwas very sensitive, but only moderately specific, in

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

DETECTION OF MENINGITIS EPIDEMICS 159

a 100

SO .

j -| 40

I 20

1979-18831983-1984

>5 >10 >15 >20 >2S

Threshold Value (cases/100 000/week)

>30

b 100

6 0 j

120

>5 >10 >1S >20 >25

Threshold Value (cases/100 000/woek)

>30

predicting annual disease rates greater than 100cases/100000. To improve the specificity of thisthreshold, this meningitis rate was averaged over 2weeks (equivalent to 30 cases/100000 per 2-weekperiod) and compared with the single-week threshold(Figure 5). As anticipated, using the weekly rateaveraged over 2 weeks improved specificity butmoderately reduced the sensitivity of detecting anepidemic compared with the 1-week threshold.

15 cases 100,000 week 1979-8315 cases 100,000 week 1983-8415 cases 100,000 week averaged over 2 weeks, 1979-8315 cases-100,000 week averaged over 2 weeks, 1983-84

100 ,

Sensitivity Specificity PositivePredictive Value

FIGURE 5 A comparison of the sensitivity, specificity and positivepredictive value of the 15 case/l000O0/week threshold and the 15case/100000/week threshold averaged over 2 consecutive weeks.

c 100

iO 80o

i 40J

> 20 -

1079-18831983-1984

>5 >10 >15 >20 >25

Threshold Value (cases/100 000/week)

>30

FIGURE 4 Use of weekly meningitis rates as thresholds to predictannual meningitis rates greater than 100 cases/lOOOOO/year in twenty-three^ x 40km2 areas, a. Sensitivity of thresholds during 1979-1933and 1983-1984. b. Specificity of thresholds during 1979-1983 and1983-1984. c. Positive predictive value of thresholds during 1979-1983and 1983-1984.

In order to be useful, the threshold for declaring anepidemic should be crossed early enough during anepidemic to allow rapid control. To determine thetimeliness of the threshold of 15 cases/lOOOOO/weekaveraged over 2 weeks, the potential number of casesprevented during the 1983-1984 season was examined(Table 1). If universal vaccination had been completedwithin 2 weeks after exceeding the threshold rate(assuming 100% vaccine efficacy and an additional 2weeks to mount an effective immune response), 1727(58.2%) cases in 1983-1984 could have been preventedby vaccinating in areas that had crossed the thresholdvalue. Using this threshold as a standard for declaringan epidemic, a total of 707 000 people would have beenvaccinated in 1983-1984 and 24.4 cases could havebeen prevented per 10000 doses of vaccine admin-istered. In comparison, use of the more sensitivethreshold (5 cases/100 000/week averaged over 2weeks) resulted in more potential cases being prevented

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

160 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 1 Meningitis cases potentially preventable if vaccinations hadbeen instituted after surpassing various biweekly meningitis incidencethresholds, 1983-1984*

Meningitisthreshold6

5 cases/100000

10 cases/100 000

15 cases/100000

20 cases/100000

25 cases/100000

Populationvaccinated (Vt)

782067(87.6)

747 353(83.7)

706927(79.2)

706927(79.2)

706927(79.2)

No. of casespotentially

prevented (%)

2246(75.8)1968

(66.4)1727

(58.2)1446

(48.8)1071

(36.1)

No. of casespotentially

prevented per10000 dosesof vaccine

28.72

26.33

24.43

20.45

15.15

* Assuming a 2-week delay after threshold is crossed for implementa-tion of a vaccination campaign and development of immunity.b Cases per 100000 population averaged over 2 weeks.

(75.8%), more cases prevented per 10000 doses of vac-cine administered (28.7 cases prevented per 10000doses), and more total vaccine distributed (782000doses).

Influence of VaccinationTo evaluate the possibility that prior vaccination mighthave influenced the incidence of meningitis during thestudy period, vaccination distribution records wereexamined for the study area. Table 2 shows the annualamount of vaccine distributed in the study area, theestimated per cent of the population vaccinated, thecumulative vaccination rate, and the correlation bet-ween disease rate and cumulative vaccination rate.Vaccination rates were low throughout the study

TABLE 2 Correlation between meningitis rate and vaccination rate intwenty-three 40 x 40km2 areas with populations exceeding 10000

Year

1979-19801980-19811981-19821982-19831983-1984

Populationvaccinated

(%)

1.65.9

16.71.44.9

Cumulativeper cent

vaccinated'

1.67.5

24.225.630.5

Spearmancorrelationcoefficient,

Rb

-0.100.63

-0.030.550.12

/•value

0.640.0010.890.010.58

"Cumulative per cent of study population vaccinated since July 1979.b Correlation between cumulative vaccination rate in each area andannual meningitis rate in each area for the twenty-three 40 x 40km2

areas.

region, with the highest rate of vaccination attainedduring a major vaccination campaign initiated in1981-1982. Generally, vaccination rates were not cor-related with meningitis rates, suggesting that the lowlevels of vaccination achieved in the study region didnot influence overall meningitis rates. In 1980-1981and 1982-1983, meningitis and vaccination rates weredirectly correlated, probably as the result of vaccineadministration in areas with high meningitis rates.

DISCUSSIONDespite the difficulties of conducting meningococcalmeningitis surveillance in sub-Saharan Africa, ourstudy indicates that surveillance can provide useful in-formation for controlling outbreaks of disease.Despite the potential for misclassification of men-ingococcal meningitis when a clinical case definition isused, surveillance data can detect the emergence of anepidemic. The reason for this is probably that Neisseriameningitidis is the most frequent cause of meningitisduring the dry season in the region7 and the onlycause of epidemic meningitis. Therefore, it is likelythat increases in meningitis incidence during our studywere primarily due to increased meningococcal diseaseactivity.

In our study of meningitis patterns in Burkina Faso,we found a clear correlation between meningitis ratesand village size during both endemic and epidemicperiods. We were unable to strictly control for urbanversus rural access to health care, but an examinationof only those villages with health centres suggests thatthis relationship is not an artifact of underreportingin rural areas. Also, because meningitis surveillancerecords were obtained directly from health-care andpublic health centres, failure to report cases to theMoPH was not a source of bias in our data. Differ-ences in meningococcal vaccination coverage for urbanand rural areas are unlikely to account for our findingssince urban dwellers tend to have higher immunizationrates than rural inhabitants in Burkina Faso (personalcommunication, David Sokal, Family Health Interna-tional). In our study, cumulative vaccination rates overthe 5-year study period were not related to the men-ingitis rate, presumably due to the low overall level ofachieved vaccination coverage.

In contrast to our results, others have not found arelationship between population size and meningitisrate in sub-Saharan Africa. Lepeyssonie8 found nocorrelation between provincial population density andmeningitis rate in Burkina Faso during meningitis epi-demics in the 1950s. However, meningitis rates at theprovincial level may not accurately reflect the intensityof disease at the village level since meningitis rates are

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

DETECTION OF MENINGITIS EPIDEMICS 161

not uniform over large areas. In a study conductedduring a group A meningococcal epidemic in TheGambia, Greenwood et al. did not find a relationshipbetween village size and meningitis rate.9 Our studydiffered from the Gambian study in that we examineda number of villages and cities with populations over2000.

It is not clear that the higher rate of meningitis inlarger villages is due to increased crowding. Crowdingwas not directly assessed in our study, and othersocioeconomic and demographic factors, such as trans-portation routes, nutrition, and sanitation, could playa role in the higher rates of disease in larger villages. Inaddition to regional population density, householdcrowding has also been studied as a risk factor forgroup A meningococcal meningitis; three studies haveshown an association with disease,'•10J1 but anotherstudy did not find this association.9 Householdcrowding may increase the risk of disease by increasingthe transmission of meningococci or coincident upperrespiratory infections."

Our findings have implications for vaccinationstrategies. Although universal vaccination of thepopulation is the preferred strategy during an out-break/ larger villages should be targeted for vaccina-tion during an epidemic if there is limited availabilityof vaccine. Furthermore, based on our finding thatmeningitis activity was not uniformly distributedthroughout the study region, it is reasonable to con-centrate vaccination efforts in those areas showingevidence of epidemic activity.12

We have also attempted to predict the occurrence ofepidemics using weekly meningitis rates. Given thehigh sensitivity and specificity of the threshold of 15cases/100 000 population/week averaged over 2 weeks,this rate appears to be an appropriate threshold forinitiating investigation and control activities in BurkinaFaso. By requiring that this level be maintained over a2-week period, the chances of unnecessarily institutingvaccination during nonepidemic periods is decreased.

Once epidemic disease has been detected, however,there is no advantage in using a stringent thresholdcutoff for instituting vaccination in adjacent areas.During an epidemic period, the prevalence of epidemicdisease in a region is high, and the predictive value ofcrossing a lower threshold is considerably improved(Figure 4c). Therefore, once epidemic disease isdetected in a given locale, it is reasonable to insti-tute vaccination at low threshold values (such as5 cases/100000 population/week) in nearby areas toprevent unnecessary delay. If this policy is imple-mented in areas adjacent to epidemic foci, a greaternumber of cases may be prevented.

In our study, we attempted to avoid wide fluctua-tions in weekly incidence rates by eliminating from ouranalysis 40 x 40km2 areas with under 10000 people.Similarly, since a few cases in a small population mayresult in an inaccurate estimate of disease activity, it isreasonable to apply these threshold rates to popula-tions of at least 30000 to 50000 people. To estimatethe potential number of cases prevented by vaccinationafter a particular threshold was crossed, we assumed avaccine efficacy and coverage of 100% for purposesof comparison. During an actual epidemic, however,lower degrees of coverage and efficacy will apply.

Our study was a retrospective review of dispensaryrecords, and therefore, our case detection is likely tobe more complete than routine meningitis surveillancefor most countries of the meningitis belt. In areas ofsub-Saharan Africa where little is known about thesensitivity of meningitis surveillance, these thresholdscan only be considered as an initial guideline forepidemic management. Additional studies are neededto determine the relevance of our findings to other set-tings, where modification of these thresholds may benecessary. Differences in both timeliness and sensi-tivity of case detection could also affect the reportedpatterns of disease in other countries. If reportingfrom health stations is incomplete or delayed, theactual meningitis activity may be much higher at anygiven time than indicated by surveillance figures.Under these conditions, the threshold of 15 reportedcases/100 000/week averaged over 2 weeks may be toohigh to detect an epidemic. Finally, since thresholdrates were determined in 40 x 40km2 areas with anaverage population of approximately 35000, addi-tional studies are needed to determine whether or notthese thresholds are applicable to larger populations,such as entire countries.

Because of a lack of laboratory facilities in the studyregion, our thresholds were based on reports of puru-lent meningitis, as is likely to be the case for mostdeveloping countries where these guidelines wouldbe appropriate. Although this will result in somemisclassification of cases, it is unlikely to cause a false-positive declaration of an epidemic since only Neisseriameningitidis causes epidemic meningitis. To ensurethat control measures are appropriate, however, itis recommended that bacteriological confirmation beobtained during the initial rapid assessment of anysuspected meningitis outbreak.4

Our study proposes quantitative threshold menin-gitis rates to facilitate epidemic disease management.The demonstration that thresholds can be used todetect an impending epidemic should encourage thecollection and analysis of meningitis surveillance data

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019

162 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

in other sub-Saharan countries. The importance oftimely and accurate reporting of meningitis cases insub-Saharan countries should be emphasized sincethese factors are critical for an early-warning systembased on threshold meningitis rates. While politicaland operational considerations will also affect men-ingitis control efforts, surveillance data have potentialto guide effective control strategies.

Epidemic meningococcal disease can be effectivelycontrolled if widespread vaccination is implementedearly in the course of an epidemic.2-3 Although the cur-rent group A polysaccharide vaccine is highly effi-cacious, epidemics in the meningitis belt cannot beprevented altogether with this vaccine. Routinechildhood administration is likely to be of limitedbenefit because of the short duration of vaccine-induced protection in children under 4 years old.1

Although the duration of vaccine protection is longerin older children and adults, routine vaccination ofthose aged 4 years and older during nonepidemicperiods is impractical for many countries and does notprovide coverage for young children—a large portionof the total population at risk. In the long term, it ishoped that a new group A meningococcal vaccine willbecome available that can provide prolonged group Ameningococcal immunity when administered to infantsduring routine immunization programmes. Until sucha vaccine is developed, strategies for the early detec-tion and rapid response to epidemic meningococcaldisease will remain important for the control of thisdisease.

ACKNOWLEDGEMENTSThe authors thank Dr David Sokal, Family HealthInternational and the staffs of the Ministry of PublicHealth, Burkina Faso and the United States Agencyfor International Development (USAID), BurkinaFaso for their contributions to this study. This study

was funded in part by a grant from the USAID AfricaBureau.

REFERENCES' Reingold A, Broome C V, Hightower A, et al. Age-specific dif-

ferences in duration of clinics] protection after vaccinationwith meningococcal polysaccharide A vaccine. Lancet 1985; U:114-18.

2 Cochi S L, Markowiu L E, Joshi D D, et al. Control of epidemicgroup A meningococcal meningitis in Nepal. Int J Epidemiol1987; 16: 91-97.

^ Binlcin N, Band J. Epidemic of meningococcal meningitis inBamako, Mali: Epidemiological features and analysis of vac-cine efficacy. Lancet 1982; U: 315-18.

4 Moore P S, Toole M J, Nieberg P, Waldman R J, Broome C V.Surveillance and control of communicable disease epidemicsin refugee populations: The case of meningococcal meningitis.Bull WHO 1991; 68: 587-%.

5 Zeng G, Thacker S B, Hu Z, Lai X, Wu G. An assessment of theuse of Baye's theorem for forecasting in public health: Thecase of epidemic meningitis in China. Int J Epidemiol 1988;17: 673-79.

6 Cvjetanovic B, Grab B, Uemura K. Dynamics of acute bacterialdiseases, epidemiological models and their application inpublic health. Bull WHO 1978; 56(Suppl 1): 81-101.

7 Sirol J, LaRoche R, Delprat J. Places des meningites e menin-gococoques dans les infections meningees. Medecine Tropicale1977; 37: 133-35.

8 Lepeyssonnie L. La meningite cerebrospinale en Afrique. BullWHO 1963; M(SoppO: 3-114.

9 Greenwood B M, Greenwood A M, Bradley A K, et al. Factorsinfluencing the susceptibility to meningococcal disease duringan epidemic in The Gambia, West Africa. J Infect 1987; 14:167-84.

10 Schwartz B, Al-Ruwais A, Ashi J A, et al. Comparative efficacy ofceftriaxone and rifampicin in eradicating pharyngeal carriageof group A Nelsseria menwgitidis. Lancet 1988; H: 1239-42.

1 ' Moore P S, Hierholzer J, DeWitt W, et al. Respiratory viruses andmycoplasma as risk factor! for epidemic group A meningococ-cal meningitis. JAMA 1990; 264: 1271-75.

12 Greenwood B M, Wali S S. Control of meningococcal infection inIhe African meningitis belt by selective vaccination. Lancet1980; 1: 729-32.

(Revised version received July 1991)

Dow

nloaded from https://academ

ic.oup.com/ije/article-abstract/21/1/155/655468 by guest on 14 O

ctober 2019