The insulin hook effect — why the Baby F 4,657 pmol/L reading is physiologically implausible
Prosecution claim
The Crown presented Baby F's insulin reading of 4,657 pmol/L (paired with low C-peptide) as proof of exogenous insulin administration. The jury was told this was diagnostic of deliberate poisoning. The Roche Cobas immunoassay reading was treated as quantitatively reliable in the range reported.
Counter-evidence
The 4,657 pmol/L reading is at a magnitude more consistent with adult attempted-suicide insulin overdose (200+ units administered) than with the prosecution's theory of a small-volume spike in a slow-running TPN bag. At this assay range the Roche Cobas is vulnerable to the 'hook effect': at C-peptide concentrations above approximately 60,000 pmol/L the assay's sigmoidal response curve produces non-linear and unreliable readings that can be misinterpreted as exogenous-insulin-dominant patterns. The Royal Liverpool clinical biochemistry laboratory was using the 2010 protocol at the time of testing — the 2012 protocol (which explicitly acknowledges the laboratory cannot diagnose exogenous insulin from the Cobas result alone) came into force later. Confirmatory mass spectrometry, which the Roche manufacturer's own guidance requires for forensic use, was not performed. The sample-handling protocol (gel tubes; delayed centrifugation; storage conditions) did not meet the forensic standard.
A reading at 4,657 pmol/L, if real, would have killed Baby F. The fact that Baby F survived is itself evidence that the reading is not what the prosecution said it was.
What the jury heard
The jury was told a high-insulin-low-C-peptide pattern is only explicable by exogenous insulin. The specific numerical value — 4,657 pmol/L — was presented as a measurement, not as a value requiring physiological cross-check. The hook effect was not explained. The manufacturer's requirement for confirmatory mass spectrometry was not a central feature of the Crown's case.
What the Panel says
The Joint Expert Witness Insulin Report on Babies F and L (May 2025) sets out, in technical detail, why the Roche Cobas screening immunoassay cannot bear the evidential weight the trial placed on it. The Panel considers the insulin evidence falls below the standard required for a murder or attempted-murder conviction.
What independent experts add
- The Roche Cobas immunoassay's sigmoidal response curve is non-linear at extremes; the hook effect distorts readings at very high C-peptide concentrations.
- Physiologically, 4,657 pmol/L insulin corresponds to quantities that would not fit the prosecution's theory of a small-volume spike in a TPN bag.
- Baby F's mother likely had gestational diabetes (publicly discussed in science4justice.nl); maternal insulin auto-antibodies can transfer to the infant and produce cross-reactive immunoassay readings for weeks.
- Child L's mother had Type 1 diabetes; the insulin auto-antibody transfer window is longer in T1 than in gestational diabetes.
- Sample-handling: gel-tube collection, delayed centrifugation, and storage conditions at the Royal Liverpool laboratory in 2015-2016 are not the heparinised red-top separated collection protocol the forensic standard requires.
- The Royal Surrey laboratory at Guildford is the UK forensic standard for insulin assay; Countess of Chester samples were sent to Liverpool's clinical biochemistry laboratory, which does not operate under the Forensic Science Regulator's code of practice.
- No TPN bags were retained or tested. The prosecution theory required that Letby had 'spiked' TPN stock accessible to all nursing staff on the unit, but no physical evidence of spiked TPN was found.
- The 2010 Liverpool protocol in force at the time of testing did not state the laboratory could not diagnose exogenous insulin; the 2012 protocol did. This matters because a forensic result requires a laboratory whose protocol is fit for forensic purpose.