We obtained complete demographic (age and neighborhood) and laboratory data for 2,897 of 3,026 participants (Figure 1). At enrollment (S1), in June 2015, 1,920 (66.3%) of 2,897 participants were IgG positive. At S2, in December 2015, 1,417 (67.2%) of 2,109 participants were IgG positive. At S3, in May 2016, 1,400 (66.5%) of 2,106 participants were IgG positive. At S4, in March 2017, 1,121 (67.9%) of 1,651 participants were IgG positive.

Our study provides data on population-based seroprevalence and rates of DENV seroconversion that may help with assessing the largely undocumented burden of dengue fever in Africa. The dengue burden in Africa has been conjectured to be similar to that in the Americas but has been largely unrecognized and masked by other illnesses with similar symptoms (21,22). In particular, the major strength of this study was that we longitudinally followed the same participants using 4 repeated serosurveys to measure the FOI of dengue in Burkina Faso.

Although IgG ELISA test results might be influenced by cross-reactivity with different flaviviruses, our estimate of seroprevalence was comparable to prevalence estimates from other studies, all tested using IgG ELISA: 61% of participants 1–65 years of age in Colombia (25); 83.1% in participants 15–19 years of age in Tahiti (26); 74% in participants in a low socioeconomic area in Recife, Brazil (27); and 68.7% in participants in Salvador, Brazil (28). The overall proportion of IgG-positive participants remained similar across the surveys, although it was highest at S4 (68%). The small increase in seropositivity despite an outbreak between S3 and S4 may have resulted from different participants being lost to follow-up at each consecutive survey. Such losses were substantial during the S1–S2 and S3–S4 intervals. The mean age of participants lost during S1–S2 was 20.7 years (95% CI 19.3–22.0) and was 23.7 (95% CI: 22.5–24.9) during S3–S4. The mean age of the participants was 22.2 years (95% CI 21.6–22.9) during S1–S2 and 21.6 (95% CI 21.0–22.2) during S3–S4. For S1–S2, the mean age was not significantly different between nonparticipants and remaining participants, but lost participants were significantly older for S3–S4; participants lost at follow-up might have tended to be IgG positive, plausibly resulting in underestimating IgG positivity at the follow-up survey. While decreasing representativeness is a limitation, we based FOI calculations on seroconversion meaning they likely would not have been substantially affected. We found apparent seroreversion from seropositive to seronegative in <3% of the paired samples: 29 during S1–S2, 14 during S2–S3; and 38 during S3–S4. However, we could not distinguish whether such cases were due to test errors or waning immunity.

Data from the baseline serosurvey in this study (part A) resulted in an annual FOI of 6.0%, although this assumes that it was constant over age and calendar time. This finding is comparable with estimates from other regions, such as Sri Lanka at 14% per year (15); Colombia at 8.7% (22); and a low socioeconomic area in Recife, Brazil, at 5.3% (25). Our model seems to capture the increase in baseline seroprevalence with age (Figure 1). However, any given age profile in a cross-sectional study can result from different combinations of incidence varying over age or over calendar time (27). Therefore, the estimated FOI from the baseline survey is subject to more limitations than FOIs estimated from the paired surveys. Nonetheless, our data support that dengue may be a bigger public health problem in Africa than is currently recognized.

From the repeat surveys (part B) for which we knew the exact duration of the intervals, we could measure the FOI more directly without assuming a constant value over age, providing a more accurate estimate of the magnitude of transmission. These estimates were higher than the 6% annual FOI based on the enrollment survey. Among 1,269 participants with IgG results from all 4 serosurveys, annual FOIs were 14% for S1–S2, 10% for S2–S3, and 20% for S3–S4. Testing of paired samples showed that 7.3% (33/455) seroconverted during S1–S2, 5.2% (23/443) during S2–S3, and 17.1% (78/455) during S3–S4. As we expected, the annualized estimates of rates were ≈2 times the simple proportions for the first 2 intervals. For the third interval (S3–S4), the annualized FOI was ≈1.5 times (12/8 months) the proportion seroconverting, because the conversion between proportions and rates is less linear for higher values.

The paired survey analyses did not require an assumption of constant FOI before the study and showed that FOI varied markedly across time within the study. In particular, a high annualized FOI (20%) was observed between S3–S4, coinciding with a major outbreak. However, even during nonoutbreak intervals, we found considerable dengue transmission; annualized FOIs ranged from 10% (in a nonrainy season, such as S2–S3) to 14% (in a rainy season, S1–S2) (28). Little comparable information from the region is available, but these rates are comparable to the 8.7% per year reported from Colombia (23).

These annual rates of infection, measured by seroconversion, may include both initial and subsequent heterotypic infection, because these are indistinguishable by IgG ELISA. However, in our analysis, the influence of past infection is minimal, because we focused on participants with IgG-negative status at enrollment for calculations of FOIs between surveys. This study was unable to distinguish participants with single or multiple infections during the study period. Furthermore, among those with multiple infections, cyclical dominance among DENV serotypes is observed in regions with better-documented dengue endemicity (29,30). However, cocirculation of multiple DENV serotypes in Burkina Faso has yet to be fully demonstrated. Thus, in our study, we calculated an overall FOI, for the totality of serotypes (20,31). Still, we recognize the need for more in-depth analyses using neutralization assays to understand serotype-specific transmission patterns in Africa.

In part B of the study, without having to assume that risk is constant over age, we found that FOI was higher in older than in younger participants, although these findings are based on small numbers, given that most people had already seroconverted at younger ages. FOI increased with age in the intervals of the nonoutbreak times (S1–S2 and S2–S3), which has also been found in Colombia (23). Our data are from IgG seroconversion rather than clinical dengue, but incidence of dengue fever in the 2016 outbreak in Burkina Faso was also higher among teenagers and young adults (4,6,32). In contrast, in the outbreak interval S3–S4, the FOI was similar across ages (13%–34%), possibly suggesting the emergence of a different serotype with little preexisting population immunity (20). Without much data on serotype-specific DENV incidence in Burkina Faso, previous studies and reports of ministry of health and World Health Organization investigations suggest that the 2016 outbreak was mainly caused by DENV-2 (4,6,32).

Myalgia during S1–S2, arthralgia during S2–S3, and fatigue during S3–S4 were positively associated with seroconversion with statistical significance. Whereas cases of seroconversion may have been associated with either mild or subclinical illnesses, participants were asked whether they had experienced these symptoms in the interval since the previous survey. These symptoms are common in dengue illness; myalgia and arthralgia were listed in the 1997 World Health Organization dengue case classifications, and fatigue or lethargy in 2009 classifications (33–35). However, because these self-reported symptoms were recorded at intervals of 6–8 months, without prompt investigation to identify a cause, we cannot conclude that any associations with seroconversion are causal.

This study is subject to several limitations. First, the generalizability of this study is limited by participants all being recruited from the urban population in the capital city. The magnitude and patterns of DENV transmission may differ in other regions of Burkina Faso, including rural areas, with different age profiles, ecologic settings, or socioeconomic conditions. Also, although a large number of participants (1,269) participated in all 4 serosurveys, a substantial number of original participants missed surveys. Those participants who missed >1 surveys were more likely to be older than those who participated in all surveys. This shortfall could have led us to underestimate prevalence.

Second, our results were based on serologic testing using IgG ELISA. Further analyses using neutralization assays are planned, but no confirmatory testing was applied to verify the IgG results. For estimating dengue FOI, serologic cross-reaction with other flaviviruses affecting the observed dengue IgG rates is a commonly reported limitation (36,37). Ultimately, this limitation could result in inaccurate seroprevalence and FOI estimates because antibodies against nondengue flaviviruses could have been detected (15,38). In particular, Zika virus has been reported in Burkina Faso (39,40), as have outbreaks of YF in 1998, 2003, and 2004 (41–43). Vaccine-induced YF antibodies could also result in interference with the specificity of IgG ELISA (38). In our data, at each of the serosurveys, <5% of participants answered that they had received YF vaccination; although subject to recall bias, this rate is much lower than the coverage rate of 85% reported by the Expanded Program on Immunization in 2007 (42). Using self-reported YF vaccination history as a proxy for YF virus seropositivity, we found no difference in IgG positivity between participants reporting and not reporting vaccination, a finding supported by others (38,44).

Despite the possible effect of cross-reactivity on dengue ELISAs, 2 studies reported a concordance level of 99% between the plaque reduction neutralization test (PRNT) and dengue IgG results (37,45). Also, when IgG ELISA results were verified by using serologic data among 13,661 participants from 13 countries to estimate dengue FOI, IgG ELISA results were confirmed by 50% PRNT in >97% of the IgG-positive samples (20). Samples from 277 healthy adults in a rural district in Malaysia were tested with the same ELISA test used in our study, and PRNT was performed on a subset of IgG-positive samples (46). Evidence of past infection was found in 75.5% (209/277) of participants and, of 96 samples randomly selected to undergo PRNT testing, all 96 were dengue-confirmed with >50% plaque reduction for DENV and showed that the detected antibodies were specific to dengue virus (46).

The sensitivity of this commercial DENV IgG indirect ELISA was reported as 99.2% and the specificity as 96.2% when compared with the hemagglutination inhibition (HI) method (47). A study compared performance of IgG ELISA to that of HI testing in a serosurvey of 327 children in Tahiti and reported sensitivity and specificity, as well as agreement between the 2 tests, to be >83% for all DENV serotypes (24). Another study reported 90.9% sensitivity and 92.9% specificity of the IgG indirect ELISA when compared with HI, also with a high correlation between the tests (48). Whereas those data are from outside of Africa, where a different and unknown composition of flaviviruses may be circulating, the results indicate a high degree of agreement between IgG ELISA and more confirmatory and better validated tests, such as HI and PRNT.

If cross-reactivity across other flaviviruses were to result in misclassifications, leading to a high rate of false positives, our findings would be overestimates of the true disease prevalence and FOI. However, given the high concordance between IgG and PRNT or HI assay results based on available data and because we analyzed data from paired serosurveys of participants who were IgG negative at enrollment, we believe that our results were minimally affected by this issue.

In conclusion, our estimates of both seroprevalence and FOI were comparable with those from dengue-endemic countries in the Americas. Repeated outbreaks indicate a considerable level of DENV transmission in Ouagadougou, but the extent of transmission and hyperendemicity needs to be further verified. Specifically, additional longitudinal evaluation with confirmatory tests and linked clinical evaluation of dengue fever in the general population in the region would be necessary to further validate our findings. Seroprevalence and FOI are important factors to be considered when making evidence-based decisions to implement interventions for prevention and control, including vaccine introduction. In the absence of other reliable data, our findings on dengue seroprevalence and FOI based on consecutive serosurveys provide practical evidence that could be used to support policy decisions.

Additional details of methodology used in study of force of infection for dengue virus, Ouagadougou, Burkina Faso, June–July 2015.