Scientifically uncertain – probability and causation in clinical negligence
Alexander v HCA International Ltd and Dr David Reffitt [2026] EWHC 1284 (KB)
The use and misuse of statistics in clinical negligence is a perennial topic, which is not surprising given that experts and lawyers must attempt to translate the statistics in relation to cohorts of patients that are set out in scientific literature to the circumstances of an individual claimant.
Back in 2020, QMLR published an article by John Whitting KC about two cases where statistical association was found not to be of itself proof of medical cause or, in a legal context, of causation.
There have been a number of recent judgments grappling with what scientific data can tell us about legal causation, and the differences between how lawyers and doctors think about probability.
In Hakmi v East & North Hertfordshire NHS Trust & Anor [2025] EWHC 2597 (KB) the court addressed odds ratios in thrombolysis for stroke; in LXLP v St George’s University Hospitals NHS Foundation Trust [2026] EWHC 560 (KB) it looked at risk ratios in relation to antibiotic prophylaxis against chorioamnionitis.
This judgment concerns a clinical negligence claim arising out of an endoscopic retrograde cholangiopancreatography (‘ERCP’) procedure performed on 16 July 2019 at London Bridge Hospital. The Claimant alleged that 100mg rectal Diclofenac should have been prescribed and administered peri-operatively to reduce the risk of post-ERCP pancreatitis, and that the failure to do so caused or materially contributed to his severe acute necrotising pancreatitis and its life-changing sequelae. Quantum had been agreed at £550,000, subject to liability.
Background
The Claimant underwent ERCP for removal of a common bile duct stone. The First Defendant owned and operated the Hospital. The Second Defendant, a consultant gastroenterologist, maintained that it was his usual practice to prescribe rectal Diclofenac in all ERCP cases to reduce the risk of post-procedure pancreatitis. It was agreed that it was the Second Defendant’s responsibility to decide what medication was required, and that it was the First Defendant’s nursing staff’s responsibility to administer the medication prescribed by the Second Defendant. However, Diclofenac was not recorded in the drug prescription and administration chart or the discharge medication summary, although it appeared in the Second Defendant’s ERCP report. Following discharge, the Claimant developed severe acute necrotising pancreatitis, requiring prolonged hospitalisation, multiple operations, intensive care treatment, and leaving him with permanent bowel dysfunction and type 3c diabetes.
Issues
The issues included whether the Second Defendant had provided a written prescription or oral instruction for Diclofenac; whether the First Defendant’s nursing staff had in fact administered it; and whether, if Diclofenac had not been given, that omission caused the pancreatitis on a “but for” basis or at least made a material contribution to it.
What there was not, was any issue as to vicarious liability on the part of the Private Hospital for the Second Defendant. It was accepted that the First Defendant owed the Claimant a duty in contract, and was vicariously liable for the Private Hospital’s clinical and nursing staff.
Breach of Duty
HHJ Richard Roberts found that the Claimant had not been informed, either at consultation or on the day of the procedure, that Diclofenac would be administered rectally. The court held that this should have been explained as part of the consent process and that the relevant box on the consent form should have been completed. The judge further found that, by virtue of Regulation 227 of the Human Medicines Regulations 2012, directions for administration of a prescription-only medicine in hospital had to be in writing. No such written prescription was given.
The court rejected the Second Defendant’s late-developed oral evidence that he had orally instructed the nurses to administer Diclofenac and that there had been a collaborative read-out of drugs given during and after the procedure. That account was found to be inconsistent with the contemporaneous documents and with his own witness statement. The judge also noted that “What was conspicuous was that the Second Defendant’s evidence was all based on his standard practice and not his recollection of the Claimant and the ERCP.” The Judge found that “the Second Defendant’s evidence from the witness box, which grew on each telling, that he instructed the First Defendant’s nurses to administer Diclofenac and Diclofenac was administered by [one of the nurses] is not true and I reject it. I find that this seriously undermines the Second Defendant’s credibility.”
The judge concluded that Diclofenac was neither prescribed, orally or in writing, nor administered. The Second Defendant had intended to prescribe it, but had did not do so and failed to communicate that instruction to the nurses. The claim against the First Defendant hospital was dismissed, because in the absence of a prescription or instruction the nursing staff were not shown to have breached any duty of care. It is interesting that by the time of trial the First Defendant had amended its Defence to accept that the Second Defendant had given the necessary instructions, and to aver that the nurses had given the diclofenac; the judge rejected this and found that the hospital’s initial position, which was a potential cut-throat defence denying that the surgeon had given such instructions to the nurses, was more likely to be true.
Causation
A substantial part of the judgment addressed whether, on the balance of probabilities, rectal Diclofenac would have prevented the Claimant’s pancreatitis.
It was accepted by the First Defendant in closing that the crucial issue was whether the “administration of rectal diclofenac is likely to have reduced the risk of pancreatitis by more than 50% in this case”.
The gastroenterology experts had agreed that the ‘relative risk ratio’ rather than the ‘odds ratio’ would be the correct statistical tool, equating to ‘balance of probabilities’ although the odds ratio and relative risk ratio would likely approximate closely.
The Second Defendant’s expert was of the view that only a 95% confidence level would reach the threshold of a balance of probabilities, objecting that “if you reduce the certainty, you’ll be able to find whatever number you want because then the range of uncertainty is so great that it becomes a meaningless statistic”. The judge rejected this and found that the court’s task was not to determine scientific certainty at a 95% confidence interval, but to decide causation on the ordinary civil standard of the balance of probabilities.
In finding that causation was made out, the court found that it was appropriate to rely on subgroup analysis of the data, focusing specifically on rectal Diclofenac, rather than broader meta-analyses involving other NSAIDs or ineffective routes of administration such as oral, intravenous, or intramuscular delivery. Despite the Defendants’ position that this weakened the statistical reliability of the analysis due to smaller numbers, the court followed the guidance of Lord Nicholls in Gregg v Scott [2005] 2 AC 176, [32] that “the value of the statistics will of course depend upon … the closeness of their position to that of the claimant” and found that the subgroup analysis gave the most valuable statistics due to their closeness to the Claimant’s circumstances, despite the Defendants’ contention that other aspects of the subgroup analysis made the Claimant’s circumstances different (e,g. whether the Claimant was at high or low risk of pancreatitis in the first place, or the amount or timing of administration of diclofenac).
The judge concluded that the studies specific to rectal Diclofenac showed a reduction in risk sufficient to satisfy the balance of probabilities test in that the relative risk was reduced below 50%.
Accordingly, had 100mg Diclofenac been administered rectally pre-operatively as intended, the Claimant would probably have avoided pancreatitis altogether.
Although unnecessary in light of that finding, the judge also held that the Claimant would in any event have succeeded on a material contribution basis, because the expert evidence established that Diclofenac would have made a more than negligible contribution to reducing the inflammatory cascade that led to pancreatitis.
Comment
The judgment is a useful reminder of the pitfalls in the use of statistics to establish or deny causation; and the potential for misunderstandings between experts and lawyers in relation to probability and the balance of probabilities. In this case the judge found that a risk ratio of less than 0.5 satisfies the balance of probabilities test, through showing a reduction in risk of more than 50%. This of course is the corollary of the test for causation where an action or delay in treatment “doubles the risk” of harm (see the analysis of Jay J in Rich v Hull and East Yorkshire Hospitals NHS Trust [2015] EWHC 3395 (QB). It is also illustrative to compare the case of Choudhary v South Central Ambulance Services NHS Trust [2015] EWHC 1311 (QB) where the court rejected the idea that the absolute risk reduction of adverse outcome was the appropriate test in respect of the balance of probabilities and accepted that the relative risk reduction was the appropriate test.
It is however worth noting that in LXLP (By her mother and litigation friend, CYLP) v St George’s University Hospitals NHS Foundation Trust [2026] EWHC 560 (KB), Kimblin J held at that “The risk ratios alone are a tool and not an answer. Paraphrasing Lord Nicholls in Gregg v Scott at [27 and 28], though the statistical evidence may be the main tool available to understand what would have happened to a claimant, prospectively, the statistics are necessarily retrospective records, and for different patients. They do not show whether the claimant would conform to trend, or not.” As this case shows, the court will very much be prepared to dive into the data and subgroup analysis in order to get to the closest match it can.
As for the way the court approached how confidence intervals affected the statistical analysis, this may illustrate a difference in understanding between clinicians and lawyers in relation to the impact of statistical confidence intervals on questions of probability; this could well be revisited in an appropriate case.
(The now) Mr Justice Cotter once remarked to one of the editors of this newsletter in court that “the problem with you lawyers is that you don’t like maths, you like logic”. It is perhaps for that reason that use and misuse of statistics remains so much at issue in clinical negligence cases.
