WITHDRAWN — 2026-09-10
This page previously served a pooled efficacy estimate for this question. That estimate has been withdrawn. The original page is preserved byte for byte and is linked below; nothing has been deleted. This notice is deliberately better sourced than the page it replaces — every trial below carries a resolvable registration and a PMID.
No pooled estimate is published for this question. The
withdrawn pool published OR 0.4941 and a derived "~51% efficacy"; an unpublished
re-derivation is shown near the end, explicitly not published and not promoted.
No published pooled estimate stands in the withdrawn one's place.
The pooled estimate OR 0.4941 (95% CI 0.3466 to 0.7045) across three trials,
and the derived headline "roughly 51% efficacy" (computed as 1 − 0.4941).
Vaccine efficacy is 1 − RR, 1 − IRR, or 1 − HR. It is never 1 − OR. This page pooled odds ratios and subtracted the pooled odds ratio from 1 to obtain an efficacy percentage. That is a category error that biases the efficacy figure.
The registry record we read states the measure in its own field, verbatim:
paramType: "Efficacy = 1-Relative Risk"
We read that field, discarded the stated measure, and computed an odds ratio instead. Two of the three trials computed efficacy from person-time incidence rates (Armah: 79/2610.6 person-years vs 129/2585.9; Isanaka: 2.14 vs 6.44 per 100 person-years) — person-time denominators this page discarded entirely.
Every contributing 2×2 table mixed a per-protocol numerator with a randomised denominator. Worse, the Rotarix cells are fabricated:
Rotarix, as served: 21 / 1042 vs 37 / 1047 1042 + 1047 = 2089 = the study-level enrollmentInfo.count, SPLIT IN TWO. The true randomised arm sizes (1647 / 1651 / 1641) sit in the participant-flow module of the same registry record, and were not used.
A study-level integer was taken and two arm denominators were invented from it. The object
underlying this page records the mismatch of its own accord — two trials carry a
provenance_MISMATCH field stating the counts do not match any posted outcome, and
Rotasiil (NCT02145000) posts no
results section at all, so its counts cannot have come from the registry this page's masthead
claims as its source.
Our served Rotarix efficacy read 43.9%. The published and registry-posted figure is 61.2% (44.0 to 73.2). The extraction error moved the pool toward the null, understated heterogeneity (true I² ≈ 71%, served 57.6%), and manufactured the caveat the page was built around.
The page said its Hartung–Knapp interval crossed no-effect because of
disagreement between the trials. That is false on the page's own arithmetic: the
variance-inflation factor is q = 0.920 < 1, so the floor bound applied and the
HK standard error equalled the unadjusted one — the entire widening was
t(2) = 4.3027 versus z = 1.96, i.e. small-k, not
between-trial dispersion. There is less dispersion than the model expects, the
opposite of what the page claimed. With the Rotarix cells corrected the crossing disappears:
OR 0.438 (0.308–0.623), Hartung–Knapp 0.202–0.949,
which excludes no-effect.
The page reported no harms at all, including intussusception — the signature safety concern for oral rotavirus vaccines. For a vaccine review that is disqualifying on its own.
Stated only so this withdrawal is not read as denying the vaccines work. The three trials' own reported efficacies, each from its publication and registration:
| trial | registration | PMID | reported VE |
|---|---|---|---|
| Rotarix Africa (Madhi 2010) — South Africa & Malawi | NCT00241644 | 20107214 | 61.2% |
| RotaTeq Africa (Armah 2010) — Ghana/Kenya/Mali | NCT00362648 | 20692030 | 39.3% |
| Rotasiil ROSE Niger (Isanaka 2017) | NCT02145000 | 28328346 | 66.7% |
A pool of the three, computed correctly, is approximately VE 56% (37 to 70%), with I² ≈ 71%, Q 7.09, p = 0.029.
The reviewer's judgement is that these should probably not be pooled at all: three different vaccines, three different settings, and follow-up windows differing roughly twofold and never disclosed on the page (which is exactly why two of the trials used person-time). This re-derivation has not cleared review; it is shown for transparency, not as a replacement claim.
Preserved unmodified at archive/ROTAVIRUS_VACCINE_AFRICA_REVIEW_2026-09-10.html.
bytes 1050240 sha256 103a4184f7e7f9798a3b4899bb5a49faaba8d9a311133ae57030c9963d2bc4a1 git blob (unfiltered) 70431c486f4c9d632366de43cf41b4b15a109505
The archived copy is byte-identical to what this URL served immediately before this notice. A withdrawal is itself evidence: a reader who followed a citation to the withdrawn estimate must be able to see exactly what was claimed, and a deletion would erase that record and break every inbound link.
Withdrawn 2026-09-10. This notice replaces the page at its canonical address; the address itself is unchanged.