We're pleased to have Krzysztof Sakrejdacome down from UMass Amherst to give a talk on using Stan for public health. Here's a description of his talk:

In recent years major players in infectious disease have become more interested in applied infectious disease forecasting. Improving predictions of flu in the U.S. would be an obvious target for these efforts. Unfortunately U.S. data on flu incidence are only available at an extremely aggregated level---four regions cover the entire country---which makes it nearly impossible to get the benefits available from spatial and hierarchical time-series models.

The Thai Ministry of Public Health (MoPH) has worked with us to create a database of snapshots of individual records of Dengue infection in its 76 provinces for the period (1968-present) and in ~8,700 sub-districts (1999-present). We update the dataset biweekly as new cases are reported making this an excellent dataset for testing infectious disease forecasting methods.

I'll talk about some challenges and successes we've had so far in processing this stream of data, constructing reasonable models, evaluating forecasts and introducing our results into decision-making. The modeling is a mishmash of time-series, survival, non-parametric methods and linear regression with a healthy serving of cross-validation. I'll focus on a Bayesian hierarchical survival-type model we are using to account for delays in case reporting and its application in the context of forecasting but I'd be happy to talk about the rest. There will be pretty maps.