OUTBREAK INVESTIGATION · TRANSPARENT ANALYSIS
Explore the pattern.
Challenge the conclusion.
From a participant list to an epidemic curve, exposure comparison and a reviewable report.
Educational prototype by Prem Kumar Perumal. Start with fictional data or de-identified participant records only. No names, addresses or patient identifiers. Calculations do not establish causation or replace an outbreak investigation.
01 / THE INVESTIGATION
Does the association survive?
This cohort workflow needs all relevant participants, including non-cases. A case-only list cannot estimate attack rates or risk ratios for the whole group.
Non-cases stay included. Cases outside the selected definition are excluded—not recoded as healthy. Unknown exposures are excluded from the 2×2 analysis, but included cases remain on the epidemic curve.
Plain CSV, no quotes, up to 1 MB / 5,000 rows. Header: id,onset,status,exposed. Exposure: 1=yes, 0=no, ?=unknown. Status: confirmed, probable, uncertain or noncase.
Inspect participant records
First 50 rows shown; report calculations use every accepted row.
02 / TIME, NOT JUST TOTALS
When did illness begin?
Counts use symptom-onset dates, not report or laboratory-result dates. Zero-count dates are shown. Bins are exploratory; choose intervals appropriate to the investigation. A shape alone does not identify the source or mode of spread.
03 / CALCULATOR
Inspect a 2×2 table.
Enter counts independently of the imported cohort. Risk calculations apply to cohort data with appropriate denominators; odds ratios can be used for case–control analysis. Case–control sampling does not estimate population attack rates.
04 / DENOMINATORS MATTER
Disease frequency.
For cumulative incidence, count new cases in a defined period among an initially at-risk population. For point prevalence, count existing cases at one point in time. For incidence rate, use total at-risk person-time and document how events are counted. No denominator adjustments, standardisation or confidence intervals are applied here.
05 / REVIEW THE EVIDENCE
Know what the numbers mean.
Risk ratio = [a/(a+b)] ÷ [c/(c+d)]. Odds ratio = ad/bc. Risk difference = exposed risk − unexposed risk. 95% intervals use log-Wald approximations, with 1.96 standard errors. These may perform poorly with sparse data; exact methods are not implemented.
With any zero cell, confidence intervals are withheld. Undefined ratios are shown as “Not estimable”; no automatic continuity correction is added. Associations are unadjusted and do not account for confounding, clustering, selection bias or multiple exposure comparisons.
The case-definition sensitivity panel repeats the same complete-case analysis under three definitions. This is not a missing-data sensitivity analysis. Unknown exposure records require investigation; treating them as unexposed would be misleading.
Beta / Demo. Not a validated surveillance platform. No automatic outbreak alerts, source attribution, geolocation, prediction or public-health notifications. Review data provenance and case definitions before interpreting results.
