BAY-States Cholera: Data Exploration & Trigger Analysis

Suspected cholera surveillance in Borno, Adamawa and Yobe (2018–2023) and its implications for an anticipatory-action trigger

Overview

What this is. An exploratory analysis of the Borno–Adamawa–Yobe (BAY) suspected-cholera line-list, 2018–2023, and what it implies for an anticipatory-action (AA) trigger. It refreshes the earlier BAY cholera exploration using the consolidated CHOLERA_BAY 2018_2023 line-list, and follows the structure of the team’s recent DRC cholera trigger analysis: characterise the data, interrogate its quality, describe the epidemiology, then test a candidate trigger and validate it against the historical CERF Rapid Response allocations.

Purpose

Northeast Nigeria carries the large majority of the country’s cholera burden and has twice drawn CERF Rapid Response funding (2018, 2021). This report asks three questions:

1. What does the surveillance line-list actually contain — how complete is it, and can it support a real-time trigger?
2. What is the epidemiology — where, when and how large are BAY cholera outbreaks?
3. Would a percentile-plus-growth trigger have worked — how often would it fire, and would it have flagged the outbreaks that drew CERF money, with useful lead time?

Bottom line

• Scale. 32,826 suspected cholera cases and 836 deaths (overall CFR ≈ 2.5%) across the three BAY states, 2018–2023.
• Three big years, three thin ones. 2018, 2021 and 2022 are large outbreak years; 2019 and 2023 are thin, and 2020 is essentially missing (3 cases) despite confirmed outbreaks that year — a data gap, not an epidemiological lull.
• A candidate trigger (weekly cases above the 99th-percentile population fraction or a ≥4× week-on-week jump, sustained 3 weeks) yields 27 LGA-year activations over the period, concentrated in the known outbreak years.
• CERF alignment. The trigger fires in both CERF years (2018, 2021) weeks-to-months ahead of the actual Rapid Response allocation dates — encouraging for lead time, but data timeliness (not shown by this retrospective file) remains the binding constraint.

This is analysis, not an endorsed framework

No BAY cholera AA framework has been published. The trigger explored here is a candidate design for discussion; thresholds and the priority-LGA set are not final. See Trigger exploration for caveats.

Data sources

Source	What	Coverage	Notes
CHOLERA_BAY 2018_2023.xlsx	Case line-list (one row per suspected case)	Borno, Adamawa, Yobe · 2018–2023	Onset date, LGA, ward, outcome, age/sex, OCV. Read from blob pa-aa-nga-cholera/raw/healthsector/.
nga-adm2-populations-2023.csv	2023 LGA populations	55 BAY LGAs	Denominator for per-capita thresholds.
CODAB nga_adm2	LGA boundaries	Nigeria	For choropleth maps.
CERF Rapid Response	Allocation dates	2018-10-31, 2021-10-22	Calibration reference events (per DS knowledge base).

Analysis structure

• 1 · Data overview — what the line-list holds, coverage by year, state and LGA, and outcomes.
• 2 · Data quality — reporting volume over time, the 2020 gap, and what it does to percentile baselines.
• 3 · Epidemiology — epidemic curves, seasonality, spatial burden, demographics.
• 4 · Trigger exploration — the candidate trigger, its components, and alert heatmaps.
• 5 · Activation frequency — how often it fires, where, and the implied return period.
• 6 · CERF validation — does the trigger flag the 2018 and 2021 CERF outbreaks, and with what lead time?

Approach

The line-list is cleaned once (state-name casing, LGA-name crosswalk to the population file, onset-date parsing) and aggregated to zero-filled LGA × epidemiological-week and state × week panels — zero-filling matters because percentile thresholds computed only on weeks with cases are far too high to be useful. All logic lives in analysis_utils.py; every chapter imports it so the numbers are identical throughout. Charts use the HDX v2 palette (Borno = blue, Adamawa = red, Yobe = amber).

Glossary

• BAY — Borno, Adamawa, Yobe states (northeast Nigeria).
• LGA — Local Government Area (admin level 2); the spatial unit here.
• Suspected case — a case meeting the clinical cholera definition in the line-list (not necessarily lab-confirmed).
• Alert week — a week where cases exceed the percentile threshold or growth exceeds the multiple.
• Sustained trigger / activation — alert level held for 3 consecutive weeks; the first such week in an LGA-year is one activation.
• CFR — case-fatality ratio (deaths ÷ cases).
• CERF RR — CERF Rapid Response, the rapid CERF funding window.
• Return period — average spacing between activations (e.g. “1-in-N years per LGA”).