EXPOSED
Most of the week is surveillance data cleaning, R/SAS regression, literature review, and writing reports and grant sections — all things current models draft at usable quality with supervision. What holds is designing studies whose confounders aren't obvious, running field outbreak investigations with case interviews and site visits, and standing behind a case definition or exposure conclusion that drives a public health order. The modal worker here is a state/local health department or academic analyst, not a field investigator, so the automatable share of the job is larger than the profession's self-image suggests.
Mixed — a routine tier and a judgment tier. A 10 reflects the split week: line-list cleaning, Poisson and Cox models in R, and drafting the background and methods sections of a grant are already model-drafted, while specifying a case definition mid-outbreak, deciding whether a cluster's excess is surveillance artifact or real transmission, and designing a cohort where the confounder is something nobody has measured yet still take a trained human — and that second set is a real third of the job, not a footnote.
Some physical or field component. 7 covers the fraction of epidemiologists who do periodic field work — door-to-door or telephone case interviews, restaurant or long-term-care facility walkthroughs, specimen and environmental sampling coordination, PPE-donned site entry during an outbreak — against a baseline of most weeks spent entirely in NEDSS/SAS at a desk.
Certification preferred, not legally required. At 5 you sit where the CIC or CPH credential is a hiring preference rather than a statute: the health officer or state epidemiologist signs the quarantine or closure order and carries the legal exposure, and nothing in your analysis requires a personally licensed signature the way a physician's or engineer's does.
Some relationship component. 9 is earned in the relationships that actually gate your data — the infection preventionist who calls you before the lab report posts, the clinic that returns your interview attempts, the CSTE and CDC counterparts who share unpublished cluster detail — but the published MMWR article or annual surveillance report itself is read for its numbers, not for who wrote it.
Exists to be accountable for ambiguous calls. 14 rests on calls that have no clean procedure and immediate consequences: setting the confirmed-versus-probable boundary when the assay is imperfect, deciding an exposure is causal enough to name a source publicly, recommending school closure or a contact-tracing scope on incomplete data, and defending that inference to press and legislators when a revision would look like error.
Has AI actually changed your work?