ED data are a window, not a census
Emergency-department data can reveal fast changes in health-care use and syndromic activity, but the population reaching an ED is selected by access, severity, behaviour, coding and service organization.
Weavidence Insights
Public health, research methods, and the systems behind reliable evidence.
Editorial writing from Weavidence on epidemiology, study design, reproducible analysis, scientific learning, and the design of the Weavidence ecosystem itself — useful whether or not you ever use a Weavidence product.
Featured
Start with a question
- What changes when artificial intelligence is used to construct the epidemiological variables that public health later analyses?
- What changes when the population at risk is estimated dynamically by a model rather than treated as a fixed count?
- How should epidemiology evaluate an AI system when its own predictions change the population it is supposed to observe?
- What happens when many public-health institutions rely on the same small set of AI models for analysis, synthesis and decision support?
- What happens when AI-generated scientific or clinical content becomes part of the data used by future public-health systems?
Topics
-
Public Health Practice
Applying evidence in real public-health and health-system settings.
12 articles -
Study Design & Epidemiology
Turning a question into a design that can actually answer it.
6 articles -
Analysis & Reproducibility
Making an analysis trustworthy, traceable, and repeatable.
7 articles -
Scientific Learning
Building the reasoning that good study design and analysis depend on.
2 articles -
Product & Engineering
How Weavidence itself is designed, governed, and built.
3 articles
All articles
Showing all 30 articles.
AI as a measurement instrument
When AI determines whether an exposure, outcome, phenotype or case is present, model error becomes epidemiological measurement error. Accuracy alone is not enough; validation must follow the variable's role in inference.
Model-dependent denominators
Public-health rates are often treated as observed facts, but AI-enabled population estimation can turn the denominator itself into a model output. The indicator then inherits the model's assumptions and uncertainty.
Performative epidemiology
AI predictions can trigger public-health actions that alter the population and the data later used to evaluate the same model. Prediction, intervention and measurement then become causally coupled.
AI monoculture in public health
Shared foundation models can create hidden dependence across institutions that appear to be making independent judgments. Public-health resilience may require epistemic diversity, not only better average model accuracy.
Recursive evidence contamination
AI-generated text, classifications and summaries can re-enter health-data and research systems as if they were ordinary observations. Without provenance, future models and analyses may learn from prior model outputs rather than from the world they are intended to represent.
ED data are a window, not a census
Emergency-department data can reveal fast changes in health-care use and syndromic activity, but the population reaching an ED is selected by access, severity, behaviour, coding and service organization.
Vaccination coverage and denominators
Vaccination coverage looks like a simple percentage, but its numerator and denominator depend on eligibility, population estimates, mobility, recording completeness and time.
When surveillance fails the decision
A surveillance system can be technically sound and still fail public health if its information arrives at the wrong time, for the wrong population, or in a form that cannot support the decision at hand.
Surveillance quality has no single optimum
Sensitivity, specificity, predictive value, timeliness, completeness, representativeness and acceptability are not independent virtues. Improving one can worsen another or increase cost.
From notification to action
A disease notification is only the first state in a longer information chain. Validation, classification, deduplication, enrichment, aggregation and interpretation determine whether a report becomes usable epidemiological information.
Vaccination trajectories
Two populations can have the same final vaccination coverage while experiencing very different periods of susceptibility. Timeliness and catch-up require longitudinal thinking.
Understanding cause-of-death data
Mortality statistics are built through certification, ICD coding and rules for selecting an underlying cause. The final tabulation is standardized, but it is not a direct transcription of biological reality.
Mortality-data quality beyond coding
Mortality-data quality starts before ICD coding. Certificate content, registration completeness, temporal consistency and fitness for purpose determine what coding can preserve.
Design the intervention first
Public-health programmes become difficult to evaluate when indicators are chosen before the causal and implementation logic of the intervention is made explicit.
Completion is not effectiveness
Activities and outputs can show that a programme operated. They cannot by themselves show that outcomes changed because of it.
Adjustment is not causality
Regression adjustment can estimate a conditional association under a specified model. It cannot decide the causal question, fix selection created by design, or determine whether a variable is a confounder, mediator or collider.
Reproducible can still be wrong
Computational reproducibility asks whether the same data and methods produce the same result. Validity asks whether the result supports the intended scientific inference.
What an audit trail should preserve
Reconstructing an analysis requires more than saving scripts and datasets. The audit trail should preserve entities, transformations, versions, exclusions, parameters, alternatives and accountable decisions.
Human decision rights in AI workflows
AI can accelerate retrieval, drafting, coding and analytical support. Decisions that define the epidemiological object, causal question, exclusions, interpretation and publication accountability require explicit human ownership.
AI speed needs more provenance
AI can multiply drafts, analyses and candidate decisions faster than humans can remember how they were produced. Acceleration increases the value of provenance.
Evaluating a surveillance system
A surveillance system is useful only when it supports its stated decisions with data that are timely, sufficiently complete, interpretable, and sustainable. Evaluation is a purpose-led assessment of trade-offs, not a single score.
How Journey builds a protocol
A finished-looking protocol and a protocol whose design reasoning survives review are different documents. Journey is built to produce the second one.
Human review as governance
"AI-powered" describes what a system uses. It does not describe who is accountable for what it produces. Weavidence treats that as a design boundary, not a caveat.
Teaching reasoning, not just facts
A completion record and evidence of demonstrated competence are different claims about a learner. Weavidence Academy is built around keeping that difference visible.
From problem to question
A public-health concern becomes researchable only after its population, exposure or intervention, comparator, outcome, and time frame are explicit — and after the question is tested for feasibility, relevance, ethics, and decision value.
Evidence of competence
Competence cannot be inferred from attendance, completion, or one successful test. It requires evidence that knowledge can be selected and applied in unfamiliar situations, demonstrated across suitable tasks, and interpreted through a defensible assessment programme.
Reproducible analytical workflows
Reproducibility requires more than saved code. The exact data state, transformations, software environment, parameters, randomness, manual decisions, and expected outputs must be identifiable — and the workflow must be independently rerunnable within legitimate access constraints.
Why traceability matters
A software artifact that looks correct and a software artifact that is inspectable and independently checkable are different claims. Traceability is what turns the first into the second.
Why Lab keeps steps separate
A single "run analysis" button that returns a chart hides more than it reveals. Lab keeps quality, computation, and review as distinct, inspectable stages instead.
Why three separate products
Academy, Journey, and Lab could have been one bundled application. They are three connected products instead, and the separation is a deliberate design choice, not an accident of scope.
No articles match these filters.
Product & engineering notes
Grounded in Weavidence's own documented architecture — a distinct kind of claim from the methodological and public-health content above.
How Journey builds a protocol
A finished-looking protocol and a protocol whose design reasoning survives review are different documents. Journey is built to produce the second one.
Human review as governance
"AI-powered" describes what a system uses. It does not describe who is accountable for what it produces. Weavidence treats that as a design boundary, not a caveat.
Teaching reasoning, not just facts
A completion record and evidence of demonstrated competence are different claims about a learner. Weavidence Academy is built around keeping that difference visible.
Why traceability matters
A software artifact that looks correct and a software artifact that is inspectable and independently checkable are different claims. Traceability is what turns the first into the second.
Why Lab keeps steps separate
A single "run analysis" button that returns a chart hides more than it reveals. Lab keeps quality, computation, and review as distinct, inspectable stages instead.
Why three separate products
Academy, Journey, and Lab could have been one bundled application. They are three connected products instead, and the separation is a deliberate design choice, not an accident of scope.