MGMIMass Gathering Medicine

← Library

AI-DRAFTED — PHYSICIAN REVIEW IN PROGRESSThe citation for this entry was checked against Crossref and PubMed, but its summary and interpretation were drafted by an AI model from the published abstract and have not yet been individually read by a physician. Read the linked source before relying on this operationally.
Paper ✓ verified

Forecasting Medical Work at Mass-Gathering Events: Predictive Model Versus Retrospective Review[1]

Zeitz KM, Zeitz CJ, Arbon P · 2005 · Prehospital and Disaster Medicine

Summary

Introduction: Mass-gathering events are dynamic and challenge traditional medical management systems. To improve the system for the provision of first aid at mass-gathering events, an evaluation of two models that assist in forecasting the number of patients presenting for first-aid services was conducted. Method: A prospective evaluation of a recurrent, mass-gathering event was undertaken comparing predicted patient presentations and ambulance transfers generated by a predictive model developed by Arbon et al and a retrospective review of seven years of historical, event data as described by Zeitz et al . Results: Patient presentation rate (per 1,000 patrons) for this event was 1.6 and the transport to hospital rate (per 1,000 patrons) was 0.07. The retrospective review closely predicted the actual overall attendance.Both methods forecast the number of patients presenting on a daily basis. The prediction proved to be more accurate, on a day-by-day basis, using the Zeitz method. Conclu

Key findings

  • Zeitz retrospective method vs actual daily presentations: r=0.65, p=0.06, median variance 7% (2002 Royal Adelaide Show).
  • Arbon model vs actual: r=0.12, p=NS, median variance 19%; over-estimated most days; predictions significantly different from observed (p<0.04).
  • Historical attendance prediction excellent: r=0.95, p<0.001 (median daily variance 8%).
  • PPR 1.6/1,000 (daily 0.9–2.5); TTHR 0.07 — double prior years' 0.034, unexplained.
  • TRANSPORTS: both methods failed — actual 47 vs Arbon 27 vs Zeitz 22, both p<0.05 from observed.
  • Paper's conclusion: the methods are complementary — Arbon when no history exists; local history wins when available.

Caveats & model limits

  • Zeitz — historical PPR (prior-event data) Requires reliable prior-event data; best with multi-year averages. Without prior presentations and attendance the model cannot run.
  • Zeitz — historical PPR (prior-event data) At a recurring show the historical method tracked observed presentations well (r=0.65, median variance 7%) and out-performed the generic model (r=0.12) - but neither reliably predicted transports.
  • Zeitz — historical PPR (prior-event data) Max daily temperature and day-of-week materially shift daily forecasts (Show presentations varied ~805-1,309 across temperature bands).
  • Zeitz — historical PPR (prior-event data) Zeitz r=0.65 carries p=0.06, median variance 7% (Arbon r=0.12, p=NS, 19%) — comparison favors history without overselling it
  • Zeitz — historical PPR (prior-event data) At the index event both methods significantly under-predicted transports (actual 47 vs 27/22, p<0.05)
  • Zeitz — historical PPR (prior-event data) Session 6: rebuilt per revised spec (BA). Historical method - computes PPR from prior-event presentations/attendance and applies it to expected attendance. Reference facts verified from Zeitz 2005 PDM PDF. | model-card caveats appended per ZEITZ_MODEL_CARD_MEMO (BA delegation, trust-batch 2026-08-08)

Read the paper

Domains

PPR & Prediction