What does AIMS actually change when an outbreak hits?
A scroll-driven simulation of one Lassa fever index case at UBTH in Benin City, comparing standard care to AIMS-equipped care across a 60-day horizon. Numbers below are central estimates from the locked methodology brief.
Without AIMS
A patient walks into UBTH with a fever. Doctors treat for malaria first, like they almost always do. Lassa is suspected only after antimalarials fail. By the time the patient is isolated, ~82 hours have passed and IPC compliance hovers around 65%.
Symptom onset → hospital presentation took ~7 days before this clock began (community-side delay; AIMS does not address this).
Total in-hospital time before isolation: — hours · ≈ 3.4 days · IPC compliance during care: —WHO IPCAF 2018, NG tertiary.
At Reff = 2.85, one index case drives a generation cycle every 13 days. Across a 60-day horizon, the spread compounds:
With AIMS
Continuous vital-sign and IPC sensors flag the syndromic pattern within the first 6-hour observation cycle. Isolation happens in ~33 hours instead of 82, and once isolated, sensor-monitored IPC keeps compliance above 90%.
Same patient, same hospital, same 7-day community delay before arrival. From here, AIMS sensors compress the in-hospital pathway.
Total in-hospital time before isolation: — hours · ≈ 1.4 days · IPC compliance during care: —Lumenix pilot deployments.
At Reff = 1.26, the same generation cycle of 13 days produces dramatically fewer cases over the 60-day horizon:
How sure are we?
No simulation has a single answer. We vary the four most-uncertain inputs to show how the headline number moves — and run the full stochastic process a thousand times to show variance the deterministic math hides.
A thousand stochastic runs of the branching process at each scenario's Reff. We report the 5th, 50th (median), and 95th percentile cumulative cases. Variance is real — a single bad index case in a poorly-IPC'd ward can generate 16+ secondaries (Fisher-Hoch 1995). The tail is in the data; we don't hide it.
What this means
One headline number, the deaths it implies at Edo's 14.2% case fatality rate, and an honest list of what this simulation does not claim.
AIMS prevents 1.59 secondary infections per Lassa index case at UBTH (central estimate, 60-day horizon). Compounded across the branching process, that translates to ~92 cases averted per index case introduced into the hospital — and approximately ~13 deaths averted at Edo's 14.2% case fatality rate (NCDC 2024).
Methodology brief, Section 7. These caveats are visible — not buried in an appendix — because the consortium audience reads them first.
- Not a clinical model. This simulates aggregate transmission, not individual patient outcomes. Real clinical decision-making is more nuanced.
- AIMS deltas are best-evidence pilot estimates, not RCT outcomes. Lumenix pilot deployments show the IPC-compliance lift and workflow time savings; we have not yet run a pre/post Lassa outcome study at a deployed site.
- Single hospital, single disease, single state. The methodology generalises; the numbers do not. A Marburg outbreak in Kano would need its own brief.
- Time-to-PCR is unchanged in the AIMS scenario. The 24-hour lab turnaround is a system constraint AIMS does not address. We name this explicitly so stakeholders don't expect AIMS to fix it.
- Patient-side delay (symptom → presentation) is unchanged. AIMS is a hospital intervention; community awareness campaigns are a separate lever.
- Branching-process variance is real. A single bad index case in a poorly-IPC'd ward can generate 16+ secondaries (Fisher-Hoch 1995). The 1,000-run Monte Carlo in Chapter 3 shows this tail; we don't hide it.