Abstract
We assessed human influenza forecasting studies to spur translation of these novel methods to practice. Searching 3 databases for papers in English, year 2000-, that validated against independent data, we included 36. They were population-based, hospital-based, and forecast pandemic spread (N=28, 4, 4, respectively); and used curve-prediction and diffusion models (N=19, 17, respectively). Four and 5 used internet search and meteorological data, respectively, besides clinical data. Eight reported sensitivity analyses; 1 compared agent-based and compartmental models. Several showed favorable 4-week-ahead skill, but lack of sensitivity analysis and model comparisons, and implementation challenges for complex models, may hinder translation to practice.