Summary of parameters varied simultaneously in sensitivity analysis and their assumed distributions.

aWe used λ = 4 in all PERT distributions, to approximate the shape of a Normal distribution. We did an additional sensitivity analysis using triangular distributions (lower bound, best estimate, upper bound) instead of PERT distributions.
bThe lower bound of the 95% predicted interval for Singapore was truncated at 1.0. Because we are dealing pooled EFT over a series of years, we would expect 1.0 to be the minimum plausible EFT. Although it is conceptually possible that EFT might be <1 for a specific region or period of time (e.g., during a dengue outbreak), the reporting in the ambulatory sector is so incomplete that while outbreaks happen periodically, we think it is conservative to assume 1 as a lower bound.
cTo be conservative, the lower bound of the PERT distribution for Malaysia is based the lower bound derived from a Delphi process in Malaysia by Shepard et al.[67]. We did not use the lower bound from the 95% CI of the regression (EFT = 5.0) because it is higher than the best estimate available (EFT = 3.8).
Notation: EFT denotes expansion factors for total dengue episodes; EFH denotes expansion factors for hospitalized dengue episodes; OP∶IP denotes outpatient to inpatient ratio; PERT denotes the distribution used in program evaluation and review technique; μ denotes mean;, σ denotes standard deviation; PI denotes prediction interval.