The underreporting pandemic: The Hidden Story Behind NHS Backlog Maintenance

One of the most revealing findings in my research was not the harm that appeared in official data, but the harm that almost certainly didn’t. Across the NHS, trusts with significant backlog maintenance issues frequently show surprisingly low levels of reported estate‑related harm. At face value, this might seem reassuring. Yet the deeper analysis — particularly the case study within my thesis — shows that low reporting does not equate to low risk. In fact, it is a sign of blind spots that prevent organisations from seeing the true impact of deteriorating estates.

The NHS relies heavily on incident reporting systems to understand patient safety trends. These systems are essential, but they are also imperfect. In fact, the Lancet argues that upto 95% of all incidents go unreported. Estate‑related harm is often subtle, indirect, or embedded within broader clinical narratives. A failing ventilation system contributes to heat stress, but the incident may be coded as a staff wellbeing issue. Water ingress might lead to a ward decant, but the disruption may be recorded under operational pressures rather than estate failure – if recorded at all. Taking an entire theatre out of action disrupts dozens of patients, but cancellations are not treated as an incident, and there is no way of linking one incident with another to get a true picture of the root cause.  As a result, many incidents that stem from environmental conditions are not recognised as such. This creates a reporting paradox: the trusts with the most fragile estates often appear to have low estate‑related harm.

The case study at the heart of my thesis illustrates this paradox clearly. The trust examined carried a substantial backlog maintenance deficit, yet reported very low levels of harm linked to estate conditions on the National Learning and Reporting System (NRLS). When viewed through traditional coding categories, the data suggested minimal risk. However, when narrative incident reports were analysed using generative AI, a very different picture emerged. The AI identified recurring equipment failures linked to environmental conditions, staff describing discomfort and heat stress, incidents concentrated in older estate zones, and repeated examples of staff creating workarounds to compensate for infrastructure deficiencies. These patterns were largely invisible in the coded dataset, yet they were present in the lived experience of staff and patients.

This disparity matters because it highlights the difference between reported harm and potential harm. Estate degradation rarely causes immediate catastrophic incidents. Instead, it creates conditions that increase the likelihood of harm over time. Unreliable equipment, compromised infection control, unsafe temperatures, reduced clinical capacity, and increased staff fatigue all contribute to a heightened risk environment. These risks accumulate and interact, amplifying each other in ways that are not captured by standard reporting systems. The low levels of reported harm therefore mask a significant potential for harm — a potential that is far greater than the data suggests.

The missed opportunity lies in how incident data is used. When estate‑related harm is under‑reported, organisations lose the ability to identify emerging risks, prioritise maintenance based on safety impact, understand how environmental conditions affect clinical outcomes, and make informed investment decisions. They also lose the chance to protect staff from avoidable strain. The case study shows that trusts could gain far richer insights by analysing narrative reports using tools such as generative AI. This approach uncovers patterns that traditional coding systems miss, revealing the latent harm that sits beneath the surface of reported incidents.

Low levels of reported harm should therefore trigger concern, not comfort. They should prompt senior leaders to ask whether staff are recognising estate‑related risks, whether incidents are being coded accurately, and whether environmental conditions are contributing to harm in ways not captured by current systems. They should also raise questions about what risks remain unseen because reporting categories are too narrow or too clinically focused. This is where the systems dynamics modelling becomes critical. The causal loop modelling shows that estate deterioration creates reinforcing feedback loops: as conditions worsen, incidents rise, capacity falls, and financial pressures increase. But if reporting fails to capture these early signals, organisations lose the chance to break the cycle.

Causal loop diagram of the impact of asset failure on the wider heathcare system

The central lesson from the case study is that the absence of evidence is not evidence of absence. Estate‑related harm is often hidden within broader clinical narratives. Without deeper analysis, organisations risk assuming safety where none exists. This is the missed opportunity: the chance to use existing data to understand risk more accurately, intervene earlier, and prevent harm before it occurs.

To address this gap, the NHS needs a couple of things. Firstly, NHS staff need to understand how to undertake concise but meaningful root cause analysis. The research highlighted the poor to non-existent levels root cause analysis within each incident report failing to understand they key factors that led to the incident. Secondly, the NHS needs a more sophisticated approach to incident analysis — one that integrates narrative analysis, AI‑driven pattern detection, estate condition data, operational performance metrics, and staff wellbeing indicators. This integrated approach would allow organisations to see the full picture of harm, not just the portion that is easily coded. It would also support more proactive investment decisions, ensuring that maintenance is prioritised based on risk rather than financial convenience.

Backlog maintenance is not simply a financial challenge. It is a patient safety challenge. The low levels of reported harm seen in many trusts are not a sign of safety; they are a sign of blind spots. The disparity revealed in the case study shows that the NHS is missing critical opportunities to understand and mitigate risk. By embracing more advanced analytical methods, organisations can uncover hidden harm, prioritise investment more effectively, and build safer, more resilient environments for patients and staff.

For some further reading on the subject, I would highly recommend the following books / articles:

The Lancet, Patient safety is not a luxury, The Lancet, 3871133

Reason, J. (1990). Human Error. Cambridge University Press. Sterman, J. D. (2000).

Business Dynamics: Systems Thinking and Modeling for a Complex World. New York: McGraw-Hill/Irwin.