A critical look at prospective surveillance using a scan statistic.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Surveillance

Background:

• The scan statistic is a widely used surveillance technique for spatial, temporal, and spatial-temporal disease data.
• It has been adapted for prospective surveillance, but its appropriateness in this context is questioned.

Purpose of the Study:

• To evaluate the suitability of the scan statistic method with standard signal rules for prospective disease surveillance.
• To assess the impact of sequential and repeated testing on the method's statistical validity.

Main Methods:

• Analysis of the scan statistic's performance under prospective surveillance conditions.
• Examination of the relationship between nominal significance level (α) and surveillance performance metrics like recurrence interval and average run length (ARL).

Main Results:

• The scan statistic does not adequately adjust for sequential and repeated testing in prospective surveillance.
• The nominal significance level (α) loses its meaning, and its relationship with recurrence interval and ARL is undefined.
• In certain scenarios, the average run length (ARL) can become infinite, rendering the method ineffective.

Conclusions:

• The scan statistic, when used with conventional signal rules, is inappropriate for prospective disease surveillance.
• The method fails to control Type-I error probability and maintain a meaningful average run length.
• Strong opposition to using the scan statistic with usual signal rules in prospective surveillance is recommended due to lack of statistical validity.