Table 2.

Comparison of alert threshold development techniques

Technique	Methods to calculate threshold [16,20]	Advantages	Disadvantages
WHO [17]	Upper third quartile of monthly case numbers from preceding 5 years	• Calculation does not require a computer • Not skewed by epidemic years	• Requires 5 years of historic data • Limited utility for local public health response when calculated at the monthly and district-wide level
Cullen [18]	Monthly mean number of cases + 2 standard deviations from 5 years of historic data where "epidemic years" have been excluded	• Simple calculation	• Requires 5 years of historic data • Limited utility for local public health response when calculated at the monthly and district-wide level • Exclusion of "epidemic years" is arbitrarily defined
C-sum [19]	Mean number of cases for a given month, the preceding month and the subsequent month from the past 5 years plus 2 standard deviations (note: the same technique has been applied to weekly data for a variety of diseases including malaria [23]).	• Smooths fluctuations due to irregular reporting rather than disease incidence by providing a larger 15 historic months sample size.	• Requires 5 years of historic data • Limited utility for local public health response when calculated at the monthly and district-wide level
Poisson [20]	Upper 95% confidence interval limit of Poisson distribution based on weekly case numbers from past 2 or more years of historic data at sites grouped by transmission zones and adjusted by population of catchment areas.	• Granular weekly and local thresholds better reflect the seasonal and geographic variations and allow for more agile public health responses • Does not require as many years of historic data • A larger historic sample size is obtained by grouping sites with similar observed patterns of transmission	• Greater influence of "epidemic years" on mean and threshold calculations because fewer years of historic data are used • Questionable applicability of Poisson assumptions • Zonal thresholds introduce an aggregation bias with inconsistent sensitivities between RHCs within a zone