The Rule That Must Justify Itself: Delta Checks as a Controlled Intervention in UK Pathology

A flag is not a neutral object

A delta check compares a current result with a previous one and asks whether the difference is plausible. In UK pathology, that question is asked thousands of times a day across analysers, sites and clinical settings. The answer is rarely a simple yes or no. A change that is unremarkable in a stable renal outpatient may be the first sign of a deteriorating inpatient; a shift that looks alarming in a general practice sample may be entirely expected in an oncology day unit.

This is why delta checks deserve more than a default percentage entered into a middleware rule. They are not passive observations. They can hold a result, trigger a repeat, prompt a phone call or release a report with a comment. Each of those actions has a cost, and each can either prevent harm or introduce delay. The central governance question is therefore not whether to use delta checks, but how to make each rule justify its place in the laboratory's control system.

Two jobs, one threshold

Delta checks are commonly asked to do two different jobs. The first is pre-analytical surveillance: catching a wrong-blood-in-tube event, a mislabelled specimen or a sample drawn from the wrong patient. The second is clinical surveillance: noticing a genuine change that may require urgent review. These jobs have different tolerances for error. A rule tuned to detect a specimen mix-up may be too sensitive for routine clinical change, while a rule tuned to clinical relevance may be too permissive for a labelling event.

Confusing the two purposes is a common source of poorly performing rules. A laboratory may inherit a delta threshold from an analyser manufacturer, apply it network-wide and then find that it either floods the duty biochemist with alerts or misses the very events it was meant to catch. Neither outcome is acceptable, but neither is surprising when the rule has never been explicitly validated against local practice.

Why generic thresholds travel badly

A single percentage threshold is a blunt instrument. Analyte behaviour varies by platform, specimen type and clinical population. Sodium, for example, is tightly regulated in health and changes slowly in most patients, but it can move quickly in critical care, in patients receiving intravenous fluids or in those with endocrine disorders. A threshold that is reasonable for a general practice population may be inappropriate for a tertiary renal service.

Analyser differences matter too. Two sites in the same pathology network may use different platforms, different reagents or different calibration arrangements. Even where methods are harmonised, local reference intervals and reporting conventions can vary. Add paediatric samples, pregnancy, haemolysis and the growing use of community point-of-care testing, and the case for a single network-wide delta threshold becomes difficult to sustain.

Designing a rule with a stated purpose

Every delta check should begin with a written intended use. What event is the rule meant to detect? What action should follow an alert? Which population and specimen type does it apply to? What time window is appropriate? These questions sound basic, but they are often left implicit, which makes later validation and audit almost impossible.

A practical design record should include the analyte, the comparison window, the threshold, the direction of change, the specimen type, the clinical setting and the expected alert rate. It should also state what the laboratory wants the alert to achieve. A rule intended to detect a possible mix-up should lead to a hold and a repeat. A rule intended to flag clinical deterioration should lead to a review of the result in context, not necessarily to a stopped report.

Validation before confidence

Validation should be proportionate to risk, but it should not be skipped. Retrospective analysis of local result data can estimate how often a proposed rule would fire and in which patient groups. That is useful for workload planning, but it does not establish that the rule detects what it claims to detect. Prospective silent running, where alerts are generated but not acted upon, can reveal operational noise before the rule is made live.

Where possible, laboratories should test rules against known or contrived pre-analytical events. Spiked samples, duplicate draws and deliberate mislabelling exercises can provide evidence of sensitivity, while review of consecutive alerts can estimate positive predictive value. The validation record should include alert burden per 1,000 results, the proportion of alerts resulting in a repeat, the proportion overridden and the time taken to resolve each alert. These are not abstract quality metrics; they determine whether a rule is usable in a real laboratory.

A network governance model

In an integrated pathology network, governance should balance consistency with local fit. A network-wide delta check formulary can set minimum standards, define common terminology and require evidence for any rule that stops a result. Local laboratories should then be able to tune thresholds within agreed limits, provided they document the rationale and audit the effect.

Each rule in the formulary needs an owner, a version number, a review date and a change-control route. When an analyser is replaced or a method changes, the rule should be revalidated rather than carried forward by default. Shared audit across sites can reveal whether a threshold performs consistently and whether local adaptations are justified. The formulary should also distinguish between rules that hold a result and rules that release it with a comment; the former carry greater operational and clinical risk and should attract stronger evidence.

Deciding when an alert should stop a result

Not every delta alert should stop a report. The decision should reflect the probability of harm, the feasibility of a repeat and the clinical consequence of delay. A suspected specimen mix-up in a high-risk analyte may justify a hold. A plausible but marked change in a patient with a known condition may justify release with a comment and a request for clinical correlation. A low-specificity alert in a stable outpatient may be better handled by a review flag than by a telephone call.

These decisions should be made in advance, not improvised at the bench. Laboratories should define escalation pathways for alerts that remain unresolved, and they should ensure that override reasons are captured in a form that supports later review. A delta alert should never become a de facto clinical decision support tool without explicit validation and clinical agreement.

Audit is the rule's second life

A delta check is not finished when it goes live. Monthly review of alert volumes, override patterns and repeat outcomes can show whether a rule is still fit for purpose. Sampling alerts for false positives and missed events is more informative than counting them alone. Feedback from clinical users can reveal whether the rule is catching meaningful changes or simply adding friction.

Review should be triggered by changes in case mix, analyser performance, clinical guidance or network configuration. The output should be a documented decision: retain, revise or retire. That discipline turns delta checks from inherited settings into controlled interventions, each with a reason to exist and a route to improvement.

The test of a threshold

A well-governed delta check does not pretend to be perfect. It states what it is for, shows evidence that it works in the local context and accepts that its performance will be reviewed. In a UK pathology service facing rising demand and increasing network working, that kind of discipline is not administrative overhead. It is the difference between a rule that protects patients and a rule that merely makes noise.