Figure 1.
Schematic diagram of multi-stream evaluation based on healthcare-seeking behaviors model for performance on outbreak detection.
First, simulated outbreak datasets are generated by the SEIR model. Second, the simulated outbreak datasets are converted into three kinds of syndromic datasets according to a discrete distribution probability of healthcare-seeking behaviors. Third, the converted syndromic datasets are superimposed onto corresponding syndromic baseline datasets to create testing datasets. Next, detection algorithms can be performed on testing datasets to detect simulated outbreaks. Last, relevant indicators can be devised to evaluate the detection performance. OTC: over-the-counter; ROC: receiver operating characteristic; SEIR: Susceptible – Exposed – Infectious – Recovered model; AMOC: activity monitoring operation curves.
Table 1.
Summary of simulated outbreaks with different parameters.
Figure 2.
Sketch diagram of conversion principle from simulated outbreak data to syndromic data.
Clinic visits data were used as an example; ai is the probability of patients visiting doctors at day i; ni is the number of new outbreak cases at day i. The number of new cases at day 1 is n1. Of these, a1n1 cases will visit doctors in clinics on that day; a2n1 cases will visit doctors on the second day; a3n1 will do so on the third day (light blue pillars), etc. On day 2, there are n2 new cases, of which a1n2 cases will visit doctors in clinics on that day; a2n2 cases will do so on the second day; a3n2 will do so on the third day (green pillars), etc. As a result, the sum of the visiting volumes on day 1 is a1n1; on day 2, it is a2n1 + a1n2; on day 3, it is a3n1 + a2n2 + a1n3; etc.
Figure 3.
Generation of semi-synthetic testing datasets in six towns in Hubei, China, 2012/4/1–2013/6/30.
A) Simulated Outbreak Data (generated by SEIR model) and converted syndromic data (generated by healthcare-seeking behaviors model); B) Testing CV Data (simulated CV + CV baseline); C) Testing OTC Frequency Data (simulated OTC + OTC baseline); D) Testing SA Data (simulated SA + SA baseline). The pink epidemic was one of the simulated outbreaks generated by the SEIR model. This could be converted into simulated clinic visits (green), simulated OTC drug purchase frequency (red), and simulated school absence (black), according to the healthcare-seeking behaviors model. The first simulated outbreak was released on 2012/4/10. Simulated syndromic data were superimposed onto corresponding baselines on the same release period (see B, C, and D). Every simulation released one outbreak to generate three testing datasets, including testing CV data, testing OTC data, and testing SA data. The simulation was repeated day by day during the whole surveillance period (2012/4/10–2013/6/30). Testing SA data on vacation breaks were defaulted as “0.” CV: clinic visits; OTC: over-the-counter; SA: school absence; SEIR: Susceptible – Exposed – Infectious – Recovered model.
Figure 4.
Baselines of three surveillance streams in six towns in Hubei, China, 2012/4/1–2013/6/30.
Gray bars show three periods separated by vacation breaks. Circles point out dips during the Chinese New Year.
Table 2.
Cross correlation coefficient between data streams in different lags.
Table 3.
Probability distribution of healthcare-seeking behaviors of residents after onset of influenza-like syndrome.
Figure 5.
Comparison of validities for all surveillance strategies using EARS ∼ 3Cs algorithms.
CV: clinic visit surveillance; EARS: the Early Aberration Reporting System; OTC: over-the-counter frequency surveillance; SA: school absence surveillance. Overall single-stream surveillance contains strategies of CV, OTC, and SA; overall multi-stream surveillance contains strategies of CV + OTC, CV + SA, OTC + SA, and CV + OTC + SA.
Figure 6.
Comparison of timeliness for all surveillance strategies using EARS ∼ 3Cs algorithm.
CV: clinic visit surveillance; EARS: the Early Aberration Reporting System; OTC: over-the-counter frequency surveillance; SA: school absence surveillance. Overall single-stream surveillance contains strategies of CV, OTC, and SA; overall multi-stream surveillance contains strategies of CV + OTC, CV + SA, OTC + SA, and CV + OTC + SA.