Reporting, investigation and just culture: why more incident data has not produced more learning

The patient-safety movement built a learning machine on two assumptions: that harmful events are under-reported, so more reports reveal more of the system, and that a serious event, once investigated, yields a removable cause. Reporting is a social act shaped by the culture it is meant to expose, so the number of reports indexes culture rather than harm, and investigation is a governance ritual as much as a learning technique. The archive grew, and the learning did not follow.

This ninth of ten units turns from the design of systems to the learning the safety movement expected from them.

The premise: an informed culture rests on a reporting culture

Reason set out the premise (Reason, 1998). A safe culture is an informed culture, which depends on an effective reporting culture, and a just culture underpins a reporting culture, in which the line between acceptable and unacceptable behaviour is clear and understood.

Reason later sharpened the asymmetry between the organisation and the sharp end (Reason, 2004): complex, well-defended systems suffer rare but catastrophic accidents when contributing factors breach successive barriers, and those in direct contact with patients have little opportunity to reform the defences. Barach and Small put the American toll at nearly 100,000 preventable hospital deaths a year, and under-reporting of adverse events at 50 to 96 per cent (Barach and Small, 2000).

Why the number of reports is not a safety measure

Vincent, Stanhope and Crowley-Murphy surveyed 42 obstetricians and 156 midwives in two obstetric units (Vincent, Stanhope and Crowley-Murphy, 1999). Almost 30 per cent could not find the list of reportable incidents, 96 per cent would always report a maternal death and fewer than 40 per cent an unexpected admission to the special care baby unit, the stated reasons being fear of blame and workload. The list is written, and a ward reading it in a second language reports only what it can.

Macrae turns that into a diagnosis (Macrae, 2015). A report also costs ward time, and a unit with no time to spare files least. A focus on the quantity of reports rather than the quality of investigation ignores the critical question of learning, since repeated reports of one event suggest a poor learning culture; it treats a biased record as a safety measure, and so a decline of one report type cannot distinguish a genuine improvement from a blind spot.

Landrigan and colleagues reviewed 2,341 admissions in ten North Carolina hospitals from 2002 to 2007, identifying 588 harms, a rate of 25.1 per 100 admissions, with no significant change in harms per 1,000 patient days (Landrigan et al., 2010).

Reporting is a social act, and that makes it readable

Waring's ethnography in a large teaching hospital shows what a report is (Waring, 2009): staff accounts of safety events are a mixture of emotion and shared notions of responsibility, and incident reports transform those accounts through check-boxes and pre-defined categories into de-contextualised narrow narratives that reflect managerial assumptions about learning.

Kaya and colleagues used NHS England staff survey data as a proxy for safety culture to predict eighteen incident-reporting variables (Kaya et al., 2023).

Griffeth and colleagues report a prospective study across nine intensive care units whose reporting had fallen below national benchmarks (Griffeth et al., 2023): a combined intervention raised reporting by 48 per cent, from 27 to 40 reports per 1,000 patient days, a rise reported as not statistically significant.

Investigation is where the learning is supposed to happen

Nicolini, Waring and Mengis spent 18 months doing ethnography in two large acute NHS hospitals (Nicolini, Waring and Mengis, 2011). Root cause analysis began as an organisational learning technique, but organisations use it as a governance tool and to re-establish legitimacy.

Peerally and colleagues make the same argument as a specification for repair (Peerally et al., 2016): root cause analysis has potential, but the field has applied it without attention to what makes it work in its contexts of origin. Organisational forgetting is widespread. The English response has changed the framework rather than the method: the Patient Safety Incident Response Framework replaced the Serious Incident Framework, required a proportionate, system-based learning response, and Louch and colleagues, reviewing 19 studies of what it replaced, found the barriers still to be time, task characteristics, blame and absent feedback (Louch et al., 2025).

The clearest evidence that investigation can fail as a practice is the Kirkup inquiry, which covered maternity and neonatal services at Morecambe Bay from 2004 to 2013 (Kirkup, 2015).

the response to adverse incidents was grossly deficient, with repeated failure to investigate properly and learn lessons (Kirkup, 2015)

The Ockenden review of maternity services at Shrewsbury and Telford Hospital NHS Trust finished with 1,486 families and 1,592 clinical incidents (Ockenden, 2022), and none of the 12 maternal deaths it considered had received care in line with best practice. External clinicians had conducted only one maternal death investigation and the reviewers rated the internal reviews poor, concluding that the service failed to investigate, failed to learn and failed to improve.

The line, and who draws it

Dekker's objection is to the form the requirement takes in every proposal he reviewed (Dekker, 2008): all argue for a clear line between acceptable and unacceptable behaviour, which alone cannot promote a just culture because it assumes culpability inheres in the act. Culpability is socially constructed, so the critical question is who gets to draw the line, not where it falls.

Dekker and Breakey set out what a just culture must supply to be more than a boundary (Dekker and Breakey, 2016): substantive justice in whether rules and procedures are fair and legitimate, procedural justice in how breaches are determined and who decides, and restorative justice in restoring the person involved and the victims.

The contested question: what would show that learning happened?

One position holds that the machinery works and its application is weak. Reason's chain is about what an organisation needs to know itself, not a claim that reports measure safety (Reason, 1998; Reason, 2004), and the deficit has addressable causes: Vincent and colleagues name clearer definitions, simplified reporting, designated staff, and education, feedback and reassurance (Vincent, Stanhope and Crowley-Murphy, 1999).

The other position holds that the machinery measures the wrong object. Macrae argues that quantity over quality ignores learning, and that declining reports cannot distinguish safety from a blind spot (Macrae, 2015); Landrigan and colleagues supply the outcome test, a decade with no significant change in the measured rate of harm (Landrigan et al., 2010); and Nicolini, Peerally and Kirkup show what happens when investigation does not convert reports into change (Nicolini, Waring and Mengis, 2011; Peerally et al., 2016; Kirkup, 2015).

The third position is the strongest counter-argument to this unit's thesis. Macrae notes that the biases are a weakness for measurement but a strength for safety management, and asks that reporting be treated as participative learning rather than data collection (Macrae, 2015), a reading Waring's finding supports (Waring, 2009). If that is right, the thesis is about how managers used numbers, not about the method.

What would settle it is a comparison, not another position statement: a change in outcome following a documented change in practice, assessed against the eight challenges of Peerally and colleagues across a sample of investigations (Peerally et al., 2016).

What to do while the dispute is open

The evidence on the two sides is unequal. That the machinery works and is applied weakly rests on a programme of theory and one survey of reporting behaviour. That it measures the wrong object rests on a decade of harm data, two ethnographies, a methods critique and two public inquiries. The strongest objection to this unit's thesis is that inquiries are commissioned because they failed, so they cannot say how often investigation works. While the dispute is open, a designer should judge a reporting system by changes in practice, not by report counts, and audit investigations against the conditions that make learning possible.

Where this is heading

Kaya and colleagues point to reporting data as one weak signal among several about culture (Kaya et al., 2023), and Macrae to participatory learning rather than data collection (Macrae, 2015). The open question is whether a reporting system will be judged on the reports it collects or on the changes it can document.

Sources

• Barach, P. and Small, S. D. (2000). Reporting and preventing medical mishaps: lessons from non-medical near miss reporting systems. BMJ. doi:10.1136/bmj.320.7237.759 Surveys what healthcare could take from near-miss reporting in other industries. It estimates nearly 100,000 preventable hospital deaths a year in the United States, with under-reporting of adverse events at 50 to 96 per cent.
• Kohn, L. T., Corrigan, J. M. and Donaldson, M. S. (2000). To Err Is Human: Building a Safer Health System. National Academy Press. doi:10.17226/9728 Reported that between 44,000 and 98,000 Americans die each year from medical errors, attributed most to system failure, and set out a national safety agenda. It is the primary behind the American toll the unit takes from a review.
• Dekker, S. W. A. (2008). Just culture: who gets to draw the line? Cognition, Technology and Work. doi:10.1007/s10111-008-0110-7 Argues that every proposal for a just culture assumes culpability inheres in the act, and that this is false because culpability is socially constructed. It asks who gets to draw the line rather than where it falls.
• Dekker, S. W. and Breakey, H. (2016). 'Just culture:' Improving safety by achieving substantive, procedural and restorative justice. Safety Science. doi:10.1016/j.ssci.2016.01.018 A conceptual paper, not an empirical study. It distinguishes substantive justice in the legitimacy of rules, procedural justice in how breaches are determined and who decides, and restorative justice in restoring the people affected.
• Griffeth, E. M. et al. (2023). Multifaceted Intervention to Improve Patient Safety Incident Reporting in Intensive Care Units. Journal of Patient Safety. doi:10.1097/pts.0000000000001151 A prospective study in nine intensive care units. A combined intervention raised reporting by 48 per cent, from 27 to 40 reports per 1,000 patient days, a rise that was not statistically significant.
• Kaya, G. et al. (2023). Exploring the impact of safety culture on incident reporting: Lessons learned from machine learning analysis of NHS England staff survey and incident data. Safety Science. doi:10.1016/j.ssci.2023.106260 Uses NHS England staff survey data as a proxy for safety culture to predict eighteen incident-reporting variables with machine learning.
• Kirkup, B. (2015). The Report of the Morecambe Bay Investigation. The Stationery Office. https://www.gov.uk/government/publications/morecambe-bay-investigation-report An independent investigation of maternity and neonatal services at Morecambe Bay from 2004 to 2013. It found 20 instances of significant or major failures of care, associated with three maternal deaths and the deaths of 16 babies, and concluded that the response to adverse incidents was grossly deficient.
• Landrigan, C. P. et al. (2010). Temporal Trends in Rates of Patient Harm Resulting from Medical Care. New England Journal of Medicine. doi:10.1056/nejmsa1004404 A retrospective review of a stratified random sample in ten North Carolina hospitals. Among 2,341 admissions it identified 588 harms, a rate of 25.1 per 100 admissions, with no significant change over time.
• Macrae, C. (2015). The problem with incident reporting. BMJ Quality and Safety. doi:10.1136/bmjqs-2015-004732 Argues that a focus on the quantity of reports rather than the quality of investigation has produced volumes carrying little new information and biased data used as a safety measure, while the bias is a strength for safety management.
• Nicolini, D., Waring, J. and Mengis, J. (2011). Policy and practice in the use of root cause analysis to investigate clinical adverse events: Mind the gap. Social Science and Medicine. doi:10.1016/j.socscimed.2011.05.010 An 18-month ethnographic study in two large acute NHS hospitals. It finds that root cause analysis was conceived as a learning technique but is also used as a governance tool and to re-establish legitimacy.
• Peerally, M. F. et al. (2016). The problem with root cause analysis. BMJ Quality and Safety. doi:10.1136/bmjqs-2016-005511 Argues that root cause analysis has potential in healthcare but has been applied without attention to what makes it work in its contexts of origin, so its potential has remained under-realised. It identifies eight challenges facing its use.
• Reason, J. (1998). Achieving a safe culture: Theory and practice. Work and Stress. doi:10.1080/02678379808256868 Sets out the argument this unit turns on: a safe culture is an informed culture, and an informed culture depends on a reporting culture underpinned by a just culture with a clearly drawn line. Culture is uniquely suited to creating the co-linear gaps in defences-in-depth through which an accident trajectory must pass.
• Reason, J. (2004). Beyond the organisational accident: the need for error wisdom on the frontline. Quality and Safety in Health Care. doi:10.1136/qshc.2003.009548 Argues that complex, well-defended systems suffer rare but usually catastrophic organisational accidents, and that healthcare shares the properties that make it subject to them. It records that those in direct contact with patients often have little opportunity to reform the system's defences.
• Vincent, C., Stanhope, N. and Crowley-Murphy, M. (1999). Reasons for not reporting adverse incidents: an empirical study. Journal of Evaluation in Clinical Practice. doi:10.1046/j.1365-2753.1999.00147.x A questionnaire study of 42 obstetricians and 156 midwives. Almost 30 per cent could not find the list of reportable incidents, and 96 per cent would always report a maternal death but fewer than 40 per cent an unexpected admission to the special care baby unit.
• Waring, J. J. (2009). Constructing and re-constructing narratives of patient safety. Social Science and Medicine. doi:10.1016/j.socscimed.2009.09.052 An ethnographic study in a large UK teaching hospital. It shows that incident reports transform those accounts through check-boxes and pre-defined categories into de-contextualised narrow narratives that reflect managerial assumptions.
• Ockenden, D. (2022). Final report of the Ockenden review of maternity services at Shrewsbury and Telford Hospital NHS Trust. Department of Health and Social Care. https://www.gov.uk/government/publications/final-report-of-the-ockenden-review Reviewed 1,486 families, 1,592 clinical incidents and 12 maternal deaths, none of which received care in line with best practice. It concludes that the service failed to investigate, learn and improve.
• Louch, G., Macrae, C., Talbot, R., McHugh, S. and O'Hara, J. K. (2025). How were patient safety incidents responded to, investigated and learned from before the Patient Safety Incident Response Framework? A rapid review. Journal of Patient Safety. doi:10.1097/PTS.0000000000001349 Reviewed 19 studies of incident response, investigation and learning in the English NHS before the framework changed, finding barriers of time, task characteristics, blame and absent feedback, and reporting that few studies measured whether learning followed.