The odds ratio was 17.6, and it still proved nothing

Between 21 July and 8 August 2017, Texas was notified of **20 cases of
cyclosporiasis** among people who had eaten at a Mediterranean-style restaurant
chain in the Houston area — four locations sharing a central kitchen where
many dishes are prepared.

The study

A case-control study with a menu-specific questionnaire focused on
fresh produce. 22 case-patients (16 confirmed, six probable) and **66
controls** — three per case.

The controls are the careful part: **dining companions of cases who did not
fall ill**, or **patrons who ate at the same location within two days of a
case's visit** — the latter traced through commercially available databases
used by Texas health agencies. Both groups ate at the same restaurant in the
same window, so the only thing left to differ is what was on the plate.

The answer

Ingredient-level analysis, matched odds ratios:

Restricted to the 16 laboratory-confirmed cases and their 48 controls,
only green onions remained significant — odds ratio 17.6 (95% CI
2.5–775.7).

Tightening the case definition eliminated three of the four candidates and
made the fourth stronger. That is what a real signal does when you remove the
noise.

And then it stopped

Restaurant invoices were collected for every item identified.

Efforts to trace any food item to its source were inconclusive.

**Investigators could not identify the source, or confirm whether these
patients had eaten a product reported by other ill people elsewhere in the
United States.**

Why this keeps happening

Two structural problems, both stated plainly:

• There is no molecular subtyping for Cyclospora. With Salmonella or
  Listeria, sequencing links a patient's isolate to a factory's swab. Here
  there is no genetic fingerprint to match, so epidemiology is the only tool
• Fresh produce has a short shelf life, and potential vehicles arrive **as
  garnishes or mixed with other items that could also carry the parasite** —
  so by the time a cluster is recognised, the food is gone and the portion that
  caused it cannot be separated from what it was served with

**The United States has had annual summer increases in cyclosporiasis since
2013**, some linked to imported produce, with past outbreaks traced to
prepackaged salad mix, raspberries and cilantro.

Which is why the conclusion is about method rather than about onions:
ingredient-level analysis within restaurant clusters remains critical —
because without a molecular link, identifying the ingredient in one subcluster
is the only evidence available about what is making people ill nationally.

Source: Centers for Disease Control and Prevention, MMWR.