We obtained from the WHO Multi-Diseases Surveillance Center in Ouagadougou annual population data and district level reports of weekly meningitis cases and deaths from Burkina Faso for the years 1996–2007 (see Table S2). We then calculated the weekly incidence and mortality, expressed as cases and deaths per 100,000 inhabitants, by dividing the number of new cases and deaths occurring per week by the mean annual population of each reporting district. A 1988 study in The Gambia found that 27 out of 154 (17.5%) survivors of bacterial meningitis had generalized neurological sequelae [28] We therefore assumed 20% of meningitis survivors would have neurological defects (sequelae).

We sorted the data into two groups: before (1996–2002) and after (2003–2007) IDSR implementation at district level. During this study period, all meningitis outbreaks in Burkina Faso only occurred between January and June (23-week period). For each group, we examined the weekly incidence rates in relation to the WHO recommended alert threshold (5 cases per 100,000) and epidemic threshold (10 cases per 100,000) [29]. We defined the start (end)of an outbreak when cases in a district exceeded (returned below) the epidemic threshold. For each outbreak, we calculated the time-to-peak of the outbreak as the number of weeks elapsed from the first alert threshold to the week with the maximum weekly incidence (i.e., the peak of the outbreak). We also calculated the time to reach the median, 25^th and 75^th percentiles of cumulative total incidence and mortality.

For each group of outbreaks before and after IDSR implementation (start 2003), we calculated the median, 25^th and 75^th percentile for each of the following health outcomes: weekly and cumulative total incidence, mortality and sequelae.

We first plotted the average weekly incidence rates over the time period studied and the median weekly incidence and mortality before and after IDSR implementation over the 23-week period of meningitis outbreaks. We then compared the health outcomes (i.e., incidence, mortality, time to peak and time to reach a set percentile of total cases per outbreak) using the Mann-Whitney test using SAS statistical software version 9.1 (SAS Institute Inc., Cary, NC, USA). In 1996 there was an “unusually” large epidemic of meningitis in Burkina Faso. We therefore examined the influence of 1996 data on the IDSR effectiveness measures by re-running the analyses excluding 1996 data.

As IDSR encompasses a deliberate response factor, it is plausible that vaccine imports may increase post-IDSR implementation. In order to assess potential correlation between IDSR implementation and meningococcal vaccine importation, we obtained estimates of the doses of vaccine imported by the Burkina Faso government from the WHO International Consultative Group, UNICEF, and GlaxoSmithKline Biologicals. Vaccine data were also collected from the WHO disease outbreak website (http://www.who.int/csr/don/en/index.html) (Tables S3 and S4). We statistically tested if annual importation of doses was impacted by IDSR implementation using the following general linear regression:Where IDSR implementation period was recorded as a dummy variable (Pre-IDSR  = 0; Post-IDSR = 1). We ran six models, each time varying the dependent variable (doses imported) as follows: Total annual doses imported (including 1996 data); Total annual doses imported (excluding 1996 data); Doses per 100,000 population in whole country (including 1996 data);

Doses per 100,000 population in whole country (excluding 1996 data); Doses per 100,000 population in districts where outbreaks occurred (including 1996 data); and, doses per 100,000 population in districts where outbreaks occurred (excluding 1996 data). To check for autocorrelation, we calculated the Durbin-Watson statistic for each regression.

We used the costs of IDSR-related activities reported in our previous study [26; see also Table S5]. The costs include those due to surveillance and response. In the case of meningitis, response is mostly treatment of those ill and vaccination of populations near a victim (e.g., those in the same village). The cost data for each activity included personnel, transportation items, office consumable goods, public awareness campaigns, laboratory and response materials and supplies, and capital items [26].

We also obtained direct medical care costs incurred by the government to treat a patient with meningitis related-illness at district health facility ($53) and regional hospital ($71) during the 2002 epidemic situation (unpublished data, Ministry of Health, Burkina Faso; see Table S6)). These estimates include only the immediate costs of meningitis case management (e.g., consultation and hospitalization fees, drugs and other essential consumables, and laboratory specimen testing). As the study perspective is the government-funded public health care system, we did not include the costs for transportation to the health facility, provision of long-term care for patients with severe sequelae and loss of productivity due to morbidity and premature death.

Assuming that any measured differences between health outcomes (i.e., after IDSR minus before IDSR) are due to IDSR, we calculated the cost-effectiveness in net cost (in dollars) per outcome averted using the following general formula:where outcomes averted were cases, deaths or sequelae.

We used the above formula to calculate the cost-effectiveness ratios for the median, 25^th, and 75^th percentiles differences in cases, deaths or sequelae averted before and after IDSR. The median (25^th, 75^th) cost effectiveness ratio was calculated using the median (25^th, 75^th) annual cost of IDSR activities In this equation, a negative result indicates net cost savings, when the cost savings of IDSR (i.e., cost of treatments avoided due to cases averted) outweigh the costs of IDSR activities (surveillance and response).

We also determined the net cost per capita of IDSR as follow:

In this equation, the per capita IDSR costs were calculated using data from the whole country [26] while the costs of treatment averted come from districts that reported outbreaks.

Regardless of IDSR activities, any measured alteration in the epidemiology of meningitis, including any reduction in cases, could have been due to naturally occurring disease cycles. To remove any unknown effect of disease cycles or specific years, we individually compared health outcomes (incidence of cases, mortality, sequelae, and time-to-peak of outbreak) between each outbreak before IDSR and each outbreak that occurred after IDSR implementation. This analysis removes the element of time sequence, and we only compare outcomes before-versus-after the

start of IDSR. Specifically, we used the following general formula:where X is a particular outbreak occurring after IDSR started in 2003, and Y is a particular outbreak before IDSR started.

Using this formula, we generated 9,030 paired comparisons (105 outbreaks before IDSR x 86 outbreaks after IDSR) in outcomes for each of the following variables: incidence of cases, mortality, sequelae, and time-to-peak of outbreak. We also examined the influence of 1996 data on these comparisons by re-running the simulation excluding all outbreaks that occurred in 1996 (giving 7,052 paired comparisons).

We then plotted the distribution of the differences between the paired comparisons (with and without the 1996 data) for the time to peak, and incidence of cases and mortality. We also calculated the median, 25^th and 75^th percentiles for the cases, deaths and sequelae averted (with and without the 1996 data). Similarly, we calculated the simple average, minimum and maximum of the differences in these health outcomes. Finally, we used the data from the paired comparisons to calculate cost-effectiveness ratios as described earlier.

Based on our definition of outbreaks (10 per 100,000 – see earlier), we identified 105 outbreaks before adoption of IDSR and 86 after adoption of IDSR (Table S2). The mean weekly meningitis cases per outbreak recorded from 1996 to 2007 at district level in Burkina Faso are represented in Figure 1. Detailed total incidence and mortality, time-to-peak of outbreak, and time to reach a fixed percentage of total incidence and mortality rates per outbreak are showed in Table S2. The size of these outbreaks was variable and 1996 was the largest epidemic year recorded (Figure 1). We observed from Figure 1 some evidence that, after IDSR implementation in 2003, there were lower peaks-of-incidence and possibly lower cumulative total incidence-per-outbreak. However, lower peak incidence after IDSR is not uniform, as we note that the peak incidence in 2007 is greater than the peaks in all but two years (1996 and 1997) before IDSR implementation.

The median and the 25^th and 75^th percentiles of the weekly number of meningitis cases and deaths before and after IDSR implementation are presented in Figures 2 and 3, respectively. The peak of the median value of the number of meningitis cases and deaths after IDSR implementation is lower than the peaks of the before IDSR median values (with and without 1996 data). However, the 75^th percentile after IDSR implementation is greater than some of the median values before IDSR, indicating that IDSR implementation does not always equate with lower number of cases (Figures 2 and 3).

The pattern of differences in peaks is also seen when comparing cumulative cases. After IDSR implementation, the average and median (50^th percentile) incidences per outbreak dropped by 135 per 100,000 and 40 per 100,000 (p-value 0.0001, Table 1), respectively. Time-to-peak, time to reach a set percentage of total cases, and time to reach a set percentage of total deaths were all reduced after IDSR; however, some p-values of the difference were greater than 0.05 (Table 1). The removal of 1996 data did reduce the size of the differences, but differences did remain, showing that IDSR implementation was correlated with reduction in incidence of cases and mortality, and time-to-peak (Table 1).

We did not find any evidence of a statistically significant correlation associated between doses of vaccine imported and outbreaks (Figure 1, Table S7). There were some obvious increases in recorded importation of vaccine in some of the years that had “large” outbreaks (e.g., 1996, 2001, and 2007). On the other hand, in some years there were little or no recorded imports, regardless of peak incidence (e.g., 1999, 2000, and 2005). We note that we had difficulty in obtaining some of the vaccine import data, particularly for the earlier years examined. We conclude that IDSR has not had an appreciable impact on doses of vaccine imported. Thus, when we considered costs associated with IDSR we did not include any costs associated with increased doses of vaccine used.

The median net cost of averting a case of meningitis was $23 per meningitis case averted (25^th percentile: $30; 75^th percentile: cost savings), and the median cost per death and sequelae averted were $98 and $126, respectively (Table 2). The median cost of operating IDSR was $0.01 per capita (25^th: $0.01; 75^th: cost saving) (Table 2). Excluding the 1996 data notably impacted most estimated cost-effectiveness ratios. For example, excluding 1996 data caused the median cost per case averted to approximately quadruple to $80 per case averted (Table 2). We also found that removing the 1996 data increased the median cost per capita to $0.02 (Table 2).

In more than 50% of the pairings in the paired outbreak analysis, which removes the influence of time and disease cycles, the outbreaks after IDSR implementation had lower incidence of cases and mortality, as well as shorter time-to-peak (Figure 4). The median, mean, maximum, and 75^th percentiles of all the paired outbreaks all showed a reduction in incidence of cases, deaths and sequelae associated with outbreaks occurring after IDSR implementation (Table 3). However, the data from paired-outbreak comparisons resulted in some lower (i.e., cheaper) median cost-effectiveness estimates (Table 4).