Real-time forecasting of an epidemic using a discrete time stochastic model: a case study of pandemic influenza (H1N1-2009)

Table 1 Estimates of parameters for the proposed model using weekly incidence data of influenza (H1N1-2009) in Japan

Week of prediction*	Initial reproduction number	Initially susceptible individuals (×10⁵)	Total number of cases (×10⁵)^†	MAE^‡
15	1.14 (0.88, 1.40)	113083 (0, 256710827)	26778 (25826, 27749)	663
18	1.18 (1.10, 1.28)	391 (218, 741)	573 (66, 1637)	1.9
21	1.15 (1.07, 1.21)	754 (0, 2225)	183 (105, 261)	0.6
24	1.15 (1.09, 1.20)	716 (540, 1104)	175 (100, 251)	0.6
44	1.13 (1.09, 1.18)	834 (664, 1149)	188 (101, 274)	0.5

The values in parenthesis are the 95% CIs. The 95% CIs for the initial reproduction number and initial number of susceptible individuals were derived from profile likelihood, while those for the total number of cases were computed using an approximate standard error of the final epidemic size. *Week by which the data were available. Using the data from week 0 to the specified week, two parameters (R _i and S ₀) were estimated and forecasts for later weeks were made. Week 44 corresponds to the end of the observation period, and the parameter estimates are based on the conditional fitting procedure using the data of the entire epidemic curve from weeks 0 to 44. ^†Estimated total number of cases from week 0 to 44, including conditionally expected values from week 0 to the week of prediction (t) and forecasting from t+1 to 44. ^‡MAE, mean absolute error; an average of absolute differences between observed and predicted values, representing a measurement of forecast error throughout the course of the epidemic.

ISSN: 1475-925X