Expert Witness Journal Issue 66 April 2026 - Flipbook - Page 19
Aggregate Data vs. Individual
Participant Data
Evidential Limits: Probability, Plausibility,
and Causation
A key distinction recognised by the court was
between AD and IPD meta-analyses. Although both
are often labelled simply as “meta-analyses”, they
di昀昀er substantially in methodological strength.
HELLP syndrome is rare, and no randomised
trials have been designed speci昀椀cally to study its
prevention. When HELLP outcomes were examined
directly in the Cochrane review, the estimated
relative risk suggested a possible reduction, but
the con昀椀dence interval crossed unity, rendering
the analysis underpowered. In statistical terms,
the available evidence could not demonstrate that
aspirin clearly prevents HELLP.
AD meta-analyses combine published summary
statistics from individual trials. While well
established, they are limited in their ability to
analyse subgroups reliably and are more vulnerable
to ecological bias, publication bias, and the
ampli昀椀cation of e昀昀ects in small subgroups.
The individual participant data meta-analysis, while
methodologically more robust, did not identify a
larger e昀昀ect or a timing-dependent bene昀椀t capable
of establishing that aspirin would probably have
prevented HELLP in this claimant. Taken together,
the highest-quality evidence suggested a possible
reduction in risk, but not one su昀케cient to satisfy the
legal test for causation.
IPD meta-analyses re-analyse raw patient-level
data across trials. This allows proper subgroup
interrogation, adjustment for confounders, and
more stable e昀昀ect estimation. Within the hierarchy
of evidence synthesis, IPD meta-analyses are
generally regarded as providing the least biased
estimates available.
The court drew a clear distinction between
evidence that indicates a trend and evidence
capable of supporting a conclusion on the balance
of probabilities. It declined to bridge that gap,
notwithstanding the intuitive appeal of the
claimant’s argument.
The court accepted that, in this context, the IPD
evidence provided a more robust and less bias-prone
basis for assessing causation, particularly where the
claimant’s case depended on timing and subgroup
e昀昀ects.
Both experts agreed that early aspirin is biologically
plausible. However, the court did not permit
biological plausibility to substitute for statistical
certainty. Plausibility explains why an intervention
might work; it does not establish that it did work,
or that it would probably have done so in this case.
This distinction is particularly important for expert
witnesses: a relative risk below 1.0, without more,
does not establish causation in an individual case,
and plausibility cannot be used to compensate for
evidential limitations. Expert witnesses should
therefore take care to separate biological rationale
from probabilistic proof, and to articulate clearly
where the evidential support for causation
legitimately stops.
Subgroup Fragility and
Population Mismatch
The claimant’s case depended heavily on subgroup
昀椀ndings relating to aspirin initiated at or before
16 weeks, and on extrapolation from severe preeclampsia outcomes to HELLP syndrome speci昀椀cally.
These 昀椀ndings arose from small subgroups within
trials that were not designed to answer those
questions.
Small subgroups can produce large relative
risk reductions that appear persuasive but are
statistically unstable. A small number of outcome
events can materially shift the point estimate, giving
an impression of certainty that is not supported by
the underlying data. The court demonstrated clear
awareness of this fragility.
Practical Lessons for Experts and Solicitors
What is particularly notable in the De Francisci
judgement, is how the court approached expert
reasoning itself. The judge examined whether the
experts’ conclusions were logically supported by
the literature they relied upon. He tested whether
assumptions were justi昀椀ed, whether extrapolations
were defensible, and whether limitations were
properly acknowledged. Cherry-picking, overreliance on fragile subgroups, and substitution
of plausibility for probability were all exposed
under scrutiny. This re昀氀ects a broader trend.
Courts are increasingly comfortable engaging with
methodological quality where causation depends on
scienti昀椀c inference.
The judgment also identi昀椀ed a signi昀椀cant
population mismatch. One of the key trials
underpinning the claimant’s analysis excluded
women with renal disease, yet the claimant had
polycystic kidney disease. This was not treated as a
peripheral technicality, but as a material limitation
on the applicability of the evidence to the individual
claimant.
For expert witnesses, this is a critical reminder that
population similarity is central to causation analysis,
even where breach is conceded.
EXPERT WITNESS JOURNAL
17
APRIL 2026
