PICO 1 Summary Report
Meningitis Outbreak Response intervention thresholds in sub-Saharan Africa
Report for the WHOMeningitis Guideline Revision
May 2014
Prepared by Dr Caroline Trotter()
Recommendation question:Following the introduction of MenAfriVac, what criteria should be used to determine when to start mass vaccination in outbreaks of meningococcal meningitis?
Current position: The WHO currently recommends for areas of population greater than 30,000: an alert threshold of 5 cases per 100,000 inhabitants per week; and an epidemic threshold of 10 per 100,000 in 1 week when epidemic risk is high, or 15 per 100,000 per week otherwise1. For small populations, thresholds are defined by absolute numbers of cases. In most instances, the operational epidemic threshold is 10 per 100,000, with the higher threshold of 15 per 100,000 being rarely used.
Background and aims
In the African meningitis belt, meningitis epidemics are detected by using weekly incidence thresholds.The current thresholds were established on the recommendation of a consensus meeting on detection of meningitis epidemics in Africa, held in Paris on 20 June 2000. Data to inform this consensus was primarily from Neisseria meningitidis group A (NmA) epidemics2-4. As the large-scale use of the NmAconjugate vaccine, MenAfriVac®, is expected to substantially reduce the burden of disease in the meningitis belt, and the epidemiology of disease due to other groups may be different to NmA,it is timely to review the current thresholds.
The aim of this paper is to address the PICO question outlined above (PICO 1). Since there have been no outbreaks of group A disease in populations immunised with MenAfriVac® and no group C or Y outbreaks have been documented in the meningitis belt in recent years, these analyses concentrate on N. meningitidis group W (NmW) outbreaks.
Methods
Data sources
Several sources of data were used to construct an NmW dataset, as summarised in table 1. There was considerable overlap in the data sources used for PICO 1 and PICO 3. All data were at district level; there were no available data at the sub-district level.
Table 1: Data sources for PICO 1 analysis
Data / Source / DescriptionSuspected case data / WHO IST Ougadougou (Clement Lingani) / Weekly case counts by district from 2005 onwards, covering most countries in the meningitis belt though not all countries for all years.
ICG vaccine requests[a] / WHO Geneva (Katya Fernandez) / Documented requests for vaccine to implement reactive immunisation campaigns 2006-2013
Laboratory line lists / WHO Geneva (Laurence Cibrelus) / Line lists of laboratory reports collated from various countries and sources
Additional data from Burkina Faso 2002, 2003 / WHO Geneva (Katya Fernandez) / Weekly case counts by district and laboratory data
Additional data from Burkina Faso 2010, 2012, 2013 / CDC (Ryan Novak) / Laboratory confirmed meningitis cases (line listing)
Additional data from Gambia 2012 / MRC Gambia (Jahangir Hossain) / Weekly case counts (suspected and confirmed) from epidemic regions with associated laboratory data
Imperial database / WHO Geneva (Katya Fernandez) / Total cases by district and year with additional laboratory data for Burkina Faso, Chad, Niger and Maliused to analyse NmAvsNmW outbreak size
Data from these different sources were incorporated into one database. Suspected case data was reorganised so that one line represented one district year with different columns showing cases by week. Laboratory line lists of individual cases were manipulated to provide totals by district and year; these were then matched to the weekly suspected case data by district and year. Additional information from other sources (table 1) was then added to this database.
District yearswith both weekly counts of suspected cases and some evidence of NmW disease were included in the NmW dataset. Evidence of NmW was usually in the form of laboratory confirmation; initially any districts with 2 or more laboratory confirmed NmW cases in a year were included. The proportion of confirmed cases that were NmW compared to all N. meningitidis confirmed cases was then examined, and district years with >50% NmW were retained. Some additional district years were included on the basis of an ICG request for NmW containing vaccine for reactive vaccination. Then, any district years with 20 or fewer suspected cases in total were excluded (33 district years).
Since surveillance is most active during the ‘meningitis season’, data from weeks 1-26 was used.
Reactive vaccination response time
Data from ICG between 2006 and 2013 was used to determine the range, mean and median time taken from a request for vaccine and implementation of a reactive vaccination campaign.
Estimating cases occurring after different weekly incidence thresholds
Thresholds of 7, 5 and 3 per 100,000 (below the current epidemic threshold of 10 per 100,000) were considered.
The week that a given threshold was crossed (wt) was identified, and the cases that occurred in subsequent weeks were summed, up to week 26 (w26).Since the seasonal incidence of meningitis is high, ‘hyperendemic’ seasonal activity may need to be distinguished from epidemic activity. Mueller & Gessner report that in Burkina Faso during January through May 2008, 96% and 79%, respectively, of the 63 districts reported a weekly incidence rate above 1 or 2 per 100,000 during at least 4 weeks5. In addition, the suspected case data may contain cases of meningitis caused by other pathogens. Therefore,in the main analyses, cases that occurred after weekly incidence declined to a ‘normal’ seasonal incidence of <2 per 100,000 (noted as wn) were excluded.
Estimating vaccine preventable cases
Because it would not be feasible to instantly implement a reactive vaccination campaign, a time lag (based on the observed reactive vaccination response time) was included, so that cases were only assumed to be prevented by vaccination following this interval (wt+lag, e.g. wt+6). The number of vaccine preventable cases was estimated by multiplying the total number of cases that occurred between wt+lag and wn by the effective vaccine coverage (VEC). The effective vaccine coverage is a composite variable that summarises both vaccine effectiveness and uptake. E.g. vaccine uptake of 95% multiplied by vaccine effectiveness of 90% gives a VECof 86%; values of 75% and 90% were used in this analysis. Some previous reactive campaigns with polysaccharide vaccine have restricted the vaccine to 2 to 29 year olds because of lower immunogenicity in young children and low disease risk in older adults. Although there is variation by outbreak, approximately 16% of NmW cases in recent outbreaks have occurred in children less than 2 years of age 6(see also PICO 3). The effect of excluding <2 year old children from the vaccine campaign was also considered, by assuming that 16% of cases occurred in this age group.The exclusion of <2 year olds in this way in the model could in practice be as a result of either not targeting this age group for vaccination or low immunogenicity in the youngest children.
Estimating deaths prevented at different weekly incidence thresholds
The number of deaths is not presented but can be estimated by applying the average case fatality experienced in NmW outbreaks (11.6%)(PICO 3 Report).
Definition of anNmW epidemic
The studies used to inform the existing thresholds defined an epidemic to be a cumulative incidence of 100 cases per 100,000 population (lower cumulative incidences of 70, 80 and 90 per 100,000 were considered in sensitivity analyses). There is evidence that NmW epidemics are, on average, less intense than NmA epidemics (Griffin et al, paper in preparation). A range of cumulative incidences are used to define an epidemic here, from a minimum seasonal incidence of 20 per 100,000 to a maximum of 100 per 100,000 (with 40, 60 and 80 per 100,000 also considered).
Threshold performance
The sensitivity, specificity, positive predictive value (PPV) and negative predictive value(NPV) of different weekly thresholds for detecting an epidemic were calculated. The definition of an epidemic season was varied between 20 and 100 per 100,000, as discussed above.
Post MenAfriVac® dataset
To investigate the properties of the thresholds further, the number of events (i.e. district years where a specific threshold was reached) that occurred in a representative dataset was estimated. Weekly suspected case data from countries that had completed MenAfriVac® campaigns was used for this purpose. This ‘post MenAfriVac® dataset’, included district years from Mali, Niger, Burkina Faso in both 2012 and 2013 and from Chad in 2013 only.
Results
Description of NmW dataset
The final dataset constructed for this analysis comprised 136 district years with both weekly suspected case data and some evidence of NmW disease. There are a total of 20,777 suspected cases, with 2318 confirmed NmW cases (11.1%confirmed overall). Burkina Faso accounted for 82(60%) of these districtyears, with Mali and Niger contributing 14 and 17 district years respectively and 7 other countries (Benin, Chad, Cote d’Ivoire, Gambia, Ghana, Guinea, Nigeria) contributing between 2 and 7 district years each. The districts included in the NmW dataset are shown in figure 1.
District population sizes ranged from 59,330 to 884,859, with a median size of 263,110. There were no districts with a population <30,000 in this dataset.
Figure 1: Map of districts in the meningitis belt with confirmed W disease between 2002 and 2013 included in this analysis. Note that some districts may be appear in the dataset for more than 1 year.
Of the 136 district years in the NmW dataset, 99 reached a cumulative seasonal incidence of 20 per 100,000, 68 district years reached 40 per 100,000, 55 district years were ≥60 per 100,000 and 36 were ≥80 per 100,000. Only 22 district-years reached the previously used epidemic definition of 100 per 100,000 and 15 of these occurred in Burkina Faso.
The total seasonal incidence ranged between 3 and 506 per 100,000 in the 136 district years. In the 99 district years exceeding a seasonal incidence of 20 per 100,000, the peak weekly incidence ranged from 2.5 to 104per 100,000 overall, with a median peak incidence of 6.2 per 100,000. Among these districts, the peak was observed between week 2 and week 17 (median week 13).
Reactive vaccination response time
There were 153 vaccine requests logged with ICG between 2006 and 2013. The mean response time from vaccine being requested to reactive immunisation being implementedwas 26 days. The minimum response time (excluding those instances where vaccine stocks were already held in-country) was 10 days.
A vaccination campaign takes 1-2 weeks to complete and a further week is required for vaccinated individuals to mount a protective immune response. The average lag time is therefore likely to be in the region of 6 weeks. We also considered an optimistic 4 week lag and an unrealistic 2 week lag for illustration purposes.
The mean time from threshold to peak weekly incidence is shown in table 2. It is clear that a lower threshold buys more time in which to respond before the peak is reached.
Table 2: Time from threshold to peak incidence
Threshold (weekly incidence per 100,000) / Number of district years reaching threshold / Mean interval from threshold to peak incidence in weeks (days)10 / 49 / 1.44 (10.1)
7 / 66 / 2.59 (18.1)
5 / 77 / 3.25 (22.8)
3 / 98 / 5.64 (39.5)
Potentially preventable cases at different weekly incidence thresholds
The number of cases occurring in the weeks after the threshold was reached, up to week 26 are shown in table 3. A more conservative count is also shown which excludes cases that occur after the incidence has returned to a normal seasonal incidence of 2 per 100,000 per week. The addition of a 6 week time lag, which seems the most likely based on ICG data, decreases the number of potentially preventable cases substantially. If a 4 week or even a 2 week time lag could be achieved, substantially more cases (approximately 2 and 3 times as many for a lag of 4 and 2 weeks respectively) are potentially preventable.
Table 3: Suspected cases occurring after weekly incidence threshold reached
Threshold (weekly incidence per 100,000) / ∑Cases week wt* to w26 / ∑Cases week wt to wn** / ∑Cases week wt+2 to w26 / ∑Cases week wt+4 to w26 / ∑Cases week wt+6 to w26 / ∑Cases week wt+6 to wn10 / 9731 / 9025 / 6951 / 4219 / 1756 / 1127
7 / 12258 / 11181 / 9170 / 5732 / 2557 / 1635
5 / 13756 / 12470 / 10895 / 7367 / 3796 / 2727
3 / 16186 / 14566 / 13891 / 10786 / 7328 / 5955
*wt= week at which the threshold is reached
** wn= week at which incidence returns to ‘normal’ seasonal baseline of 2 per 100,000 per week
More detail is given on the number of cases occurring from 6 weeks after the threshold was reached (wt+6) until return to normal seasonal activity (wn) in table 4, together with the average number of cases per district and the range.
Table 4: Number of cases occurring 6 weeks after the threshold was reached until return to ‘normal seasonal activity’ of 2 per 100,000 per week.
Threshold (weekly incidence per 100,000) / Number of district years reaching threshold / Cases occurring in weeks wt+6 to wn / Mean cases per district (range) / Median cases per district (IQR)10 / 49 / 1127 / 23* (0-434) / 0 (0, 14)
7 / 66 / 1635 / 25 (0-434) / 6 (0, 18)
5 / 77 / 2727 / 35 (0-783) / 10 (0, 31)
3 / 98 / 5955 / 61 (2-1769) / 14 (0, 67)
* The current threshold is 10 per 100,000. The number of cases occurring per event after this threshold was reached is shown for information, but vaccination was instigated at this point in many of the districts which will have curtailed the epidemic, which may make this threshold seem less favourable.
A greater number of cases (and cases per event) are prevented as the thresholds are lowered.The proportion of districts where more than 20 cases are potentially preventable increases from 24% to 35% and then to 45% as the threshold is lowered (from 7 to 5 to 3 per 100,000 respectively).
However, as the threshold is lowered, successively more individuals would have been targeted for reactive immunisation, i.e. an additional 4.0 million with a threshold of 7, an additional 7.0 million with a threshold of 5 per 100,000 and an additional 13.8 million with the lowest threshold of 3 per 100,000 (assuming the whole district population is targeted).
To investigate the robustness of these results, the distribution of the cases averted by outbreak was examined. A large number (1769) of the additional cases prevented by the lowest threshold of 3 per 100,000 per week were due to one district year (figure 2); Pissy in Burkina Faso 2002 where there was a large NmWepidemic.However, the proportion of cases occurring in this district compared to the total over all districts was similar for thresholds of 3,5 and 7 (27%), so the relative merits of the thresholds are unchanged if this district is excluded.
Figure 2: Distribution of cases occurring after a given threshold until return to normal seasonal incidence by district year
Of the district years reaching a threshold of 7 per 100,000 per week, 74% went on to pass a threshold of 10 per 100,000 per week; this was 63% for a threshold of 5 per 100,000 per week and 50% for the lowest threshold of 3 per 100,000 per week. To investigate any residual effects of vaccination triggered by the current threshold of 10 per 100,000, the districts known to have been vaccinated with an NmW-containing vaccine were excluded (table 5). The mean cases per district, i.e. those that were potentially preventable, were higher than in table 4, but the relative advantage of the lowest threshold remained.
Table 5: Number of cases occurring 6 weeks after the threshold was reached until return to ‘normal seasonal activity’ of 2 per 100,000 per week, excluding vaccinated districts where reactive campaigns with an NmW containing vaccine was implemented.
Threshold (weekly incidence per 100,000) / Number of district years reaching threshold / Cases occurring in weeks wt+6 to wn / Mean cases per district (range) / Median cases per district (IQR)10 / 32 / 1050 / 33 (0-434) / 11 (2,48)
7 / 49 / 1387 / 28 (0-434) / 14 (6, 40)
5 / 60 / 2267 / 38 (0-783) / 19 (8, 47)
3 / 81 / 5215 / 64 (2-1769) / 42 (10, 112)
Vaccine-preventable cases at different thresholds and vaccine assumptions
The estimated number of cases that could be prevented by reactive vaccination at each threshold under varying assumptions of effective vaccine coverage is shown in table 6. The exclusion of children under 2 years of age substantially reduces the number of cases that could be prevented. In all of the scenarios considered, the average number of cases prevented per event is fewer than 60, with 11 out of 16 scenarios preventing fewer than 30 cases per event.
Table 6: Cases prevented by reactive vaccinationwith different thresholds under different assumptions of effective vaccine coverage (VEC), assuming a 6 week lag
Threshold (weekly incidence per 100,000) / Cases occurring in weeks wt+6 to wn (number of events) / Cases prevented VEC=75% (per event) / Cases prevented VEC=90% (per event) / Cases prevented VEC=75%, <2y/o excluded (per event) / Cases prevented VEC=90%, <2y/o excluded (per event)10 / 1127 (49) / 845 (17) / 1014 (21) / 710 (14) / 852 (17)
7 / 1635 (66) / 1226 (19) / 1472 (22) / 1030 (16) / 1236 (19)
5 / 2727 (77) / 2045 (27) / 2454 (32) / 1718 (22) / 2062 (27)
3 / 5955 (98) / 4466 (46) / 5360 (55) / 3752 (39) / 4502 (46)
Improving reactive vaccination response
The gains in the number of cases that could be prevented if the time between threshold and effective vaccination were 4 weeks rather than 6 weeks are shown in table 7. As expected, many more cases are prevented with a shorter lag. The number of cases prevented per event is higher (better) under the current threshold of 10 per 100,000 per week if a 4 week lag is assumed than the lowest threshold of 3 per 100,000 per week with a 6 week lag.
Table 7: Cases prevented by reactive vaccination with different thresholds under different assumptions of effective vaccine coverage (VEC), assuming a 4 week lag
Threshold (weekly incidence per 100,000) / Cases occurring in weeks wt+4 to wn (number of events) / Cases prevented VEC=75% (per event) / Cases prevented VEC=90% (per event) / Cases prevented VEC=75%, <2y/o excluded (per event) / Cases prevented VEC=90%, <2y/o excluded (per event)10 / 3549 (49) / 2662 (54) / 3194 (65) / 2236 (46) / 2683(55)
7 / 4696 (66) / 3522 (53) / 4226 (64) / 2958 (45) / 3550 (54)
5 / 6181 (77) / 4636 (60) / 5563 (72) / 3894 (51) / 4673 (60)
3 / 9312 (98) / 6984 (71) / 8381 (86) / 5867 (60) / 7040 (72)
Threshold performance
The performance of the weekly thresholds compared to different definitions of an epidemic is shown in table 8. The appropriateness of the threshold is associated with the definition of an ‘epidemic’; i.e. lower thresholds are more appropriate when a lower cumulative incidence is used to define an epidemic. The ‘best’ threshold for each definition of an epidemic is highlighted. The analysis was repeated for Burkina Faso only and for all others excluding Burkina Faso, although this did not markedly change the results (not shown).
Table 8: Performance of different weekly thresholds compared to a cumulative seasonal incidence of 20, 40, 60, 80 or 100 per 100,000 population – full NmW dataset
Seasonal incidence per 100,000 / Weekly threshold per 100,000 / Sensitivity % (95% CI) / Specificity % (95% CI) / PPV %(95% CI) / NPV %
(95% CI)
20 / 10 / 49.5 (41.1, 57.9) / 100 (100,100) / 100 (100,100) / 42.5 (34.2, 50.8)
7 / 66.7 (58.7, 74.6) / 100 (100,100) / 100 (100, 100) / 52.9 (44.5, 61.3)
5 / 76.8 (69.7, 83.9) / 97.3 (94.6, 100) / 98.7 (96.8, 100) / 61.0 (52.8, 69.2)
3 / 96.0 (92.6, 99.3) / 91.9 (87.3, 96.5) / 96.9 (94.0, 99.8) / 89.5 (84.3, 94.6)
40 / 10 / 70.6 (62.9, 78.3) / 98.5 (96.5, 100) / 98.0 (95.6, 100) / 77.0 (69.9, 84.0)
7 / 94.1 (90.1, 98.1) / 97.1 (94.2, 99.9) / 97.0 (94.1, 99.9) / 94.3 (90.4, 98.2)
5 / 100 (100, 100) / 86.7 (81.1, 92.4) / 88.3 (82.9, 93.7) / 100 (100, 100)
3 / 100 (100, 100) / 55.9 (47.5, 64.2) / 69.4 (61.6, 77.1) / 100 (100, 100)
60 / 10 / 81.8 (75.3, 88.3) / 95.1 (91.4, 98.7) / 91.8 (87.2, 96.4) / 88.5 (83.2, 93.9)
7 / 100 (100,100) / 86.7 (80.7, 92.2) / 83.3 (77.1, 89.6) / 100 (100,100)
5 / 100 (100,100) / 72.8 (65.4, 80.3) / 71.4 (63.8, 79.0) / 100 (100,100)
3 / 100 (100,100) / 46.9 (38.5, 55.3) / 56.1 (47.8, 64.5) / 100 (100, 100)
80 / 10 / 91.7 (87.0, 96.3) / 84.0 (77.8, 90.1) / 67.4 (59.5, 75.2) / 96.6 (93.5, 99.6)
7 / 100 (100,100) / 70.0 (62.3, 77.7) / 54.6 (46.8, 62.9) / 100 (100,100)
5 / 100 (100,100) / 59.0 (50.7, 67.3) / 46.8 (38.4, 55.1) / 100 (100,100)
3 / 100 (100,100) / 38.0 (29.8, 46.2) / 36.7 (28.6, 44.8) / 100 (100,100)
100 / 10 / 100 (100,100) / 76.3 (69.2, 83.5) / 44.9 (36.5, 53.2) / 100 (100, 100)
7 / 100 (100,100) / 61.4 (53.2, 69.6) / 33.3 (25.4, 41.3) / 100 (100,100)
5 / 100 (100,100) / 51.8 (43.4, 60.1) / 28.6 (21.0, 36.2) / 100 (100, 100)
3 / 100 (100,100) / 33.3 (25.4, 41.3) / 22.4 (15.4, 29.5) / 101 (100,100)
Number of events occurring at different thresholds post-MenAfriVac®