34Monitoring changes in infectious disease incidence is fundamental to outbreak detection and 35 response, intervention outcome monitoring, and identifying environmental correlates of 36 transmission. In the case of dengue, little is known about how consistently surveillance data track 37 disease burden in a population over time. Here we use four years of monthly dengue incidence 38 data from three sources -population-based ('passive') surveillance including suspected cases, 39 'sentinel' surveillance with 100% laboratory confirmation and complete reporting, and door-to-40 door ('cohort') surveillance conducted three times per week -in Iquitos, Peru, to quantify their 41 relative consistency and timeliness. Data consistency was evaluated using annual and monthly 42 expansion factors (EFs) as cohort incidence divided by incidence in each surveillance system, to 43 assess their reliability for estimating disease burden (annual) and monitoring disease trends 44 (monthly). Annually, passive surveillance data more closely estimated cohort incidence (average 45 annual EF=5) than did data from sentinel surveillance (average annual EF=19). Monthly passive 46 surveillance data generally were more consistent (ratio of sentinel/passive EF standard 47 deviations=2.2) but overestimated incidence in 26% (11/43) of months, most often during the 48 second half of the annual high season as dengue incidence typically wanes from its annual peak.
49Increases in sentinel surveillance incidence were correlated temporally (correlation coefficient = 50 0.86) with increases in the cohort, while passive surveillance data were significantly correlated at 51 both zero-lag and a one-month lag (0.63 and 0.44, respectively). Together these results suggest 52 that, rather than relying on a single data stream, a clearer picture of changes in infectious disease 53 incidence might be achieved by combining the timeliness of sentinel surveillance with the 54 representativeness of passive surveillance.Infectious disease surveillance in developing countries is often challenged by limited 57 public health resources, insufficient laboratory capacity, and incomplete reporting [1]. In order to 58 obtain high-quality data in the face of these and other challenges, the World Health Organization 59 (WHO) has recommended sentinel surveillance for many infectious diseases [2,3]. In sentinel 60 surveillance systems, resources are focused on collecting complete, timely data from a subset of 61 healthcare facilities or laboratories [4], thus requiring fewer resources than would be needed to 62 actively collect the same quality of data from all facilities (population-based active surveillance). 63 Passive surveillance systems, in which data collection is dependent on reporting by healthcare 64 facilities, are representative by virtue of being population-based, but are also subject to under-65 detection and underreporting [5]. The goal of this study is to evaluate the public health utility of 66 sentinel surveillance compared to passive surveillance for measu...