Table 1 Methodologies applied to complete the four key processes of numerical weather prediction, and proposed equivalents in health prediction

Observe

Observed variables include wind, temperature, humidity, pressure, radiation, cloud cover,.. taken by different observation platforms (from land stations to satellites)

Observed variables include agents of disease (pathogens); health state of individuals and communities (hosts); environmental characteristics and triggers (host–pathogen interactions); collective phenomena inducing anomalous fluctuations

Interpret and model

The scientific relies on the laws of physics applied to a rotating fluid. The modelling is based on the numerical integrations of finite-approximations of the fluid equation. The models include parameterisation schemes that simulate most relevant processes, including stochastic terms designed to simulate, to the best of our k knowledge, processes that are poorly known, or that would be computationally too expensive to be treated explicitly

Health modelling (HM) is often multidimensional/multifactorial and targets typically the disease agent, the individual background conditions and the interactions mediated by genetic, socio-economic, lifestyle, legislative and environmental factors. HM is a complex multifaceted process that depends on interpreting and understanding contexts from which to acquire knowledge by gathering, assimilating and elaborating data and information to match complexity. Scientific understanding depends on the cross-linking between mathematical model formalism with biological and behavioural structures and dynamics

Estimate and predict

Estimates of the actual state of the system are generated by solving optimisation problems that use as input all available observations and model forecasts. Predictions of the future states are generated by integrating numerically the system equations on super computers. Ensemble methods are used to provide confidence intervals and uncertainty estimations

Estimates are subject to spatio-temporal dynamics evolving rapidly and thus requiring incremental modelling schemes that a) efficiently update the values driving health trajectories based on newly captured information and b) encompass evidence within context-driven scores and predictions to limit the general uncertainty

Diagnose and verify

Diagnostic studies are based on key events and on a sound statistical analyses of the model performance over many cases (say at least few hundred cases). Verifications of forecast quality rely on a range of metrics that provide relevant information to the forecast users and the model developers

The more data are available, the better it is. Diagnostics presents constrains (availability of tests, limited capacity etc.) that might affect model performance or at least explain the role of spatio-temporal factors in determining accuracy and reproducibility. Verification depends on the choice of metrics that must consider many data latencies and gaps and should be selected to fit multiple specific objectives, purposes and trade-offs (ethical principles, improving accessibility, reducing costs, connectivity, spatial aggregation, similarity and dissimilarity system-wide and at localized contexts etc.)