A deviation bar chart for detecting dengue outbreaks in Puerto Rico.

Abstract

OBJECTIVES: A Centers for Disease Control and Prevention deviation bar chart (Statistical Software for Public Health Surveillance) and laboratory-based surveillance data were evaluated for their utility in detecting dengue outbreaks in Puerto Rico. METHODS: A significant increase in dengue incidence was defined as an excess of suspected cases of more than 2 SDs beyond the mean for all 4-week periods from April through June (the period of lowest seasonal incidence), 1989 through 1993. An outbreak was defined as a cumulative annual rate of reported dengue greater than 3 per 1000 population. RESULTS: Retrospective application of the system to 1994 data showed agreement with previous analyses. In 1995 and 1996, 36.4% and 27.3%, respectively, of municipalities with a significant increase in reports for 2 or more consecutive weeks before the first week of September had an outbreak, compared with 9.0% (in 1995, P = .042) and 6.0% (in 1996, P = .054) of towns without a significant increase. The system showed sensitivity near 40%, specificity near 89%, and accuracy in classifying municipalities near 84%. CONCLUSIONS: This method provides a statistically based, visually striking, specific, and timely signal for dengue control efforts.

Sharing experiences: towards an evidence based model of dengue surveillance and outbreak response in Latin America and Asia

“…Our findings show that most countries did not distinguish between a seasonal rise in dengue cases, usually during the rainy season (see Figure 1), and the unexpected increase in cases above a defined threshold, usually called an outbreak [2,12]; the number of reported cases exceeding expected levels is referred to as “aberrations” [13]. The need for dengue control and clinical care systems to respond differently to each of these scenarios was identified.…”

“…Possible disadvantages of such a definition were also identified: a) Programme managers may be tempted to use the crossing of the +2SD line as a “warning sign” for an outbreak (rather than as an indicator that an outbreak is effectively already underway) and initiate a delayed emergency response; b) the definition has limited sensitivity (only 40% of such events when case numbers crossed the +2SD threshold were followed by a “massive” increase of cases in Puerto Rico [13]; using a similar predictor with a 1 SD threshold, Barbazan et al [14] found a sensitivity of 66%) c) outbreaks in previous years can result in thresholds that are too high; no satisfactory algorithm for recognising past aberrations has been devised [12]; d) the seasonal increase in cases may come earlier than in the 5 preceding years providing the impression of an outbreak. This phenomenon has been handled by using the “deviation bar chart” [13] or the “moving average” applied in Brazil and Malaysia (“moving median”).…”

“…This phenomenon has been handled by using the “deviation bar chart” [13] or the “moving average” applied in Brazil and Malaysia (“moving median”). In this case, the average of dengue cases during 4 weeks is compared with the average of cases during a period of 12 weeks in the preceding 5 years (i.e.…”

The reappearance of dengue-3 and a subsequent dengue-4 and dengue-1 epidemic in Puerto Rico in 1998.

“…Significantly increased reporting (epidemic threshold) was defined, for the island and for each municipality, as a number of case reports greater than the mean plus two standard deviations in five non-epidemic years. 15 Comprehensive vector surveillance is not routinely conducted in Puerto Rico. For inspections of case homes and neighborhoods, the Breteau index (number of containers with larvae per 100 houses inspected), the house index (number of houses with containers with larvae per 100 houses), and ovitraps were used to monitor mosquito infestation levels.…”

Assessment of dengue risk in relief workers in Puerto Rico after Hurricane Georges, 1998.

“…The epidemic threshold for each municipality was defined as 2 standard deviations above the mean incidence during months of low transmission (April, May, and June) for 5 previous nonepidemic years. 13 The incidence of dengue illness among study participants was calculated as the number of serologically positive dengue cases divided by the total number of person-days during which participants had an opportunity to develop anti-dengue antibody. The exact, one-sided upper 95% confidence interval for dengue incidence in study participants was calculated on the basis of the Poisson distribution as upper 95% ϭ Ϫln (0.05)/t, where t is person-days of exposure.…”