The Ledger of Harm: Reading NHS Patient Safety Incidents as a Record of Reporting

A fall on a hospital ward, a medication given at the wrong dose, a pressure ulcer found on admission: each may become a row in an NHS patient safety incident dataset. It is tempting to read those rows as a simple count of harm. They are not. They are the visible end of a chain that begins with someone noticing an event, deciding that it should be reported, completing a form and assigning a category. The count is real, but it measures reporting as much as it measures injury.

What an incident report actually is

In England, NHS providers record patient safety incidents through local risk-management systems and submit data to the national Learn from Patient Safety Events (LFPSE) service. An incident is not only an event that caused harm. It can be a near miss, an unexpected event that could have caused harm, or an event that was intercepted before it reached a patient. A medication error may be recorded even if the patient took no wrong dose. A fall may be recorded even if there was no fracture.

This is a strength of the system, not a flaw. Near misses are useful signals. But it means that a rising incident count cannot be read automatically as rising harm. It may mean that staff are reporting more readily, that a new digital form has made reporting easier, or that a trust has begun to classify a category more consistently. The same logic applies in reverse: a low count may reflect a quiet service, a small caseload, or a reporting culture in which people do not feel safe to speak up.

Severity coding adds another layer. Incidents are commonly graded from no harm or near miss through low, moderate and severe harm to death. Initial coding may take place before an investigation is complete. A fall first recorded as no harm may later be linked to a fracture. A pressure ulcer may be coded differently depending on whether it is judged avoidable or unavoidable, and on how firmly that judgement can be made. The dataset is a living record, not a fixed verdict.

The reporting gap is part of the measurement

Under-reporting is often discussed as if it were a scandal. Sometimes it is. More often it is a property of any reporting system. Not every incident is noticed. Not every noticed incident is reported. Staff may be busy, uncertain whether an event meets the threshold, or worried about how a report will be received. The design of the form matters too: a short, well-integrated digital route encourages reporting; a cumbersome process suppresses it.

The Patient Safety Incident Response Framework (PSIRF), used across the NHS in England, has pushed providers towards system-based learning rather than a fixed target for investigating a set number of incidents. That shift is important for readers of data. It signals that the purpose of reporting is to learn, not to produce a clean performance table. But it does not remove variation. Trusts differ in size, case mix and services. An acute hospital with many older inpatients will see different fall and pressure-ulcer patterns from a community trust or a mental health provider. A trust with more beds will usually record more incidents simply because more care is delivered.

From count to rate, carefully

Responsible analysis moves from counts to rates, but rates are only as good as their denominators. Falls are sometimes expressed per 1,000 bed days; medication incidents per 1,000 admissions or prescriptions; pressure ulcers per 1,000 admissions or per 10,000 bed days. Each denominator answers a different question. A trust with a high number of short admissions may look different under one measure than another.

Case mix also matters. A ward caring for frail, older patients with cognitive impairment will not have the same fall risk as a day-surgery unit. A specialist centre may attract patients with complex medication regimens. Comparing raw incident numbers between trusts without adjustment is therefore weak evidence. Even adjusted comparisons need caution, because coding practice is not uniform. One trust may include a broad range of near misses under 'medication error'; another may reserve the category for events that reached the patient.

What the categories conceal

Incident categories are broad containers. 'Falls' can include a patient lowered safely to the floor and a fall resulting in a hip fracture. 'Pressure ulcers' can include skin damage present on admission and damage that developed in care. 'Medication errors' can include prescribing, dispensing, administration and monitoring. The label tells you the family of events, not the severity, cause or preventability of each one.

The digital form itself shapes the data. Drop-down menus, mandatory fields and default options guide what gets recorded. That is useful for consistency, but it can also flatten nuance. A free-text description may hold the detail that explains why an event happened; the coded category may not. For anyone reading national or trust-level figures, the lesson is to ask what sits behind the label. A count of incidents is a count of records, not a count of identical clinical events.

Reading safety headlines without blame

Patient safety incident data is at its most useful when it is read for themes and trends: recurrent falls in a particular ward type, medication errors associated with a specific class of drug, pressure damage linked to delayed mobility. It is at its least useful when it is turned into a league table of blame. A higher reported count may indicate a more open reporting culture. A lower count may indicate a quieter service or a quieter reporting system. Neither inference is safe on its own.

This does not mean the data should be treated as beyond criticism. Reporting gaps are real, and they matter most where harm is severe. The point is that a report is an act of noticing before it is a measure of harm. In 2026, as NHS organisations continue to digitise incident records, the quality of the ledger depends on the quality of the reporting culture behind it. Readers should look for context: the reporting period, the severity mix, the denominator, the case mix and whether the figures are provisional. A rise in reports may be a rise in attention. A fall may be a fall in speaking up. The ledger of harm is also a ledger of how well the system is listening.