Why This Matters
Since 2023, COVID-19 testing has dropped sharply. Official case counts no longer
reflect the true spread of the virus. Wastewater
surveillance, however, is an unbiased indicator of community infection levels —
it does not depend on whether people get tested.
By calibrating current wastewater levels against historical periods when both
wastewater and case data overlapped, we can estimate how many infections are likely
occurring today, and how many of those go unreported.
How these numbers are produced
-
Two models, one range: we fit the relationship between
wastewater and cases on the calibration period
(2023-06-01 – 2024-03-01, 39 weeks).
A proportional model (zero baseline) gives
~16,734 cases/week at 2022-level testing.
A log-log (power-law) model — which allows the relationship
to curve — gives ~27,663. The average is
~22,198.
-
Testing adjustment (2022 → 2023-24): in 2022, reported cases
averaged ~57,292/week. In the calibration window,
~7,658/week — a
7.5× drop. This drop reflects both
reduced testing and lower real infection prevalence.
Note: the 7.5× factor converts estimates from
2023-24 testing levels to 2022 testing levels. The headline
(~22,198) is then ~12.6× higher
than today's reported cases (1,767) because testing has fallen
further since 2023-24 (see assumption β below).
-
Officially reported: ~1,767/week
(latest: Oct 25, 2026).
The gap (~20,431/week) represents cases not captured by testing.
Why two models?
The proportional model assumes cases scale linearly with wastewater through the
origin. The log-log model allows a curved (power-law) relationship, which is
typical for epidemiological signals. Both are fit on the same
39 weeks of overlapping data. They agree within a factor of ~1.7,
giving us confidence in the 16,734 – 27,663 range.
We do not use direct averaging over "similar weeks" because only 3 weeks in the calibration period fall near
today's wastewater level — too few for a headline estimate.
A linear regression with a free intercept gives
~36,222, but 76% of it comes
from the fitted intercept rather than the wastewater signal,
so we do not report it.
Key assumptions & limitations
-
Assumption α — 2022 → 2023-24 testing multiplier
(7.5×):
the multiplier assumes the entire drop in reported cases between
2022 and 2023-24 is due to reduced testing. If real infections were also
lower in 2023-24 than in 2022, the true multiplier is smaller and the
estimate shrinks proportionally. This is an upper-bound
assumption.
-
Assumption β — 2023-24 → today testing decline:
the model predicts 2,237–3,698 cases
at today's wastewater level if testing were still at 2023-24 levels.
We actually observe 1,767. We attribute this gap to a
further decline in testing (~1.3–2.1×, implied by the total
ratio of ~12.6× vs today's reported).
If instead the model overestimates at low wastewater levels — and
testing has not fallen further — then the true number is closer to
11,000–17,000 rather than ~22,198.
-
The wastewater→infection relationship is assumed stable,
but it was fitted on data from the 2023-24 testing regime and applied to
today's (possibly lower) testing regime. If the relationship itself shifted
as testing fell — for example, if only severe cases were tested in 2023-24
but only hospitalised cases are tested now — the model may not transfer
cleanly.
-
Week-to-week noise is substantial. The correlation on first
differences is only 0.47, which is why we report a range
(16,734 – 27,663) rather than a single figure.
-
What this is not: a case count. It is a trend and
scale indicator derived from calibrated models. Do not use it for clinical
or policy decisions without independent validation.