What is clustering, anyway?

EPID 684
Spatial Epidemiology
University of Michigan School of Public Health


Jon Zelner
[email protected]
epibayes.io

Agenda

  • What is a hotspot 🔥, anyway?.

  • Discussing the idea of kernel smoothing and how it relates to Tobler’s 1st Law.

  • Hands-on with some smoothing models.

On your own

Spend a few minutes answering these questions:

  • Based on the readings, how would you describe what a disease cluster is?

  • When might you undertake a cluster study? Is there a difference between a disease cluster and a hotspot ?

  • What are some challenges or pitfalls of cluster analysis?

05:00

Why might one undertake a cluster-based study?

  • Public Health Response.

  • Hypothesis testing.

  • Identifying key causal mechansisms.

Reasoning about clusters is tricky

The Texas Sharpshooter 🎯 Problem is an example of a logical fallacy that can come up in the analysis of spatial clusters.

What is a hotspot?

A hotspot is often in the eye of the beholder…

More detail is necessary to make the hotspot concept useful

We recommend that the meaning of a “hotspot” be made explicit by use of an appropriate modifier such as: “burden hotspot,” to denote areas of elevated disease prevalence or incidence; “transmission” or “risk hotspot,” to denote areas of elevated transmission efficiency or a higher risk of disease acquisition; and “emergence hotspot,” to denote areas with an increased probability of disease emergence or reemergence. (Lessler et al., 2017)

A cluster can represent multiple facets 💎 of risk

  • Spatial aggregation of a disease outcome.

  • Clustering of a specific exposure risk.

  • Evidence of social processes that concentrate disadvantage and multiple risks.

Smoothing!

Smoothing lets us separate signal 📡 from noise 📣

  • Often used to describe or uncover clusters.

  • What are some examples of a spatial signal or information?

  • What are some possible sources of noise that could obscure the signal?

  • What are some risks associated with mistaking noise \(\to\) signal?

A guided tour of smoothing approaches

Next Time

Time to meet John Snow & co