Richard Jones

A recent interesting study looking at AI tools to diagnose lung cancer highlights that AI does not change diagnosis speed. However, the care pathway was not changed and perhaps the most obvious finding is that care pathways must be optimised if AI is to highlight cases where specialists should take a second look. A large NHS trial found that using AI to flag abnormal chest X-rays for faster review did not meaningfully speed up lung cancer diagnosis overall. It did shorten the time for radiologists to report X-rays, but delays later in the pathway, such as CT scans, clinic appointments and follow-up processes, meant patients were not diagnosed sooner. The study analysed 93,326 chest X-rays across five NHS trusts and identified 558 lung cancer cases. Median time to diagnosis was 44 days with AI prioritisation versus 46 days without, which was not significant. Referral rates, treatment start times and cancer stage at diagnosis were also similar between groups. Researchers said the main problem is not image reporting but the wider NHS pathway. They highlighted one especially important finding: patients whose X-rays were flagged by AI but not by radiologists had much longer waits for diagnosis, suggesting this group may deserve closer study. The authors conclude that AI alone is not enough—improving outcomes would require redesigning the full care pathway so an AI alert triggers rapid follow-up actions like CT booking and specialist review.

Here is a real example from the US of why embedding patient safety can be so difficult. We assume that patient safety is something everyone cares about. But what happens when it goes up against cost imperatives? Patient safety is easiest to move forward, particularly with the Centers for Medicare & Medicaid Services (CMS) Transforming Episode Accountability Model (TEAM) initiative, when improved outcomes and safety equal cost reductions. However, even this is not a guarantee. For example: in one provider, a trial on an AI analytics package was done on a hospital and results showed, according to their own cost estimates (not the vendor's), a potential 10-20 million US dollars savings that would recur if they remained 'under control'. A 'no brainer' right? Clinicians liked it. A patient safety genius there (I'm labelling his abilities correctly) loved it. So why didn't it happen? There is no line item in the accounts for cost reduction. The finance team refused to believe it. They were under huge pressure and did not want to put their heads above the parapet so an accounting quirk led to no savings. This was potentially hundreds of millions of dollars of saving, demonstrable improvements in outcomes and protection against outside scrutiny and criticism... It still didn't happen. I'd like to say there is a happy ending. There isn't. There is a lesson. Engage all stakeholders in discussions and then, perhaps, you might make a bit more progress. However, institutional issues are going to continue to create havoc until outcomes are aligned. If revenue versus cost is the main metric (and it is in some provider systems), you'll continue to get strange decisions driven by potentially perverse incentives.

Fascinating information in this graphic. What gets measured gets improved, but a 2024 Health Services Safety Investigations Body (HSSIB) investigation revealed that systematic underreporting of patient safety incidents involving general practitioner online consultation tools was occurring, and that the available data did not contain enough information to identify potential harm. From my own direct experience, unless you have risk-adjusted metrics for patient outcomes, the layer of incidents that are not flat out Never Events also remain hidden at scale. Patient safety work is still mainly at the tip of the iceberg!

As trusts consider clearing the waiting list, there is an absence of objective approaches to prioritisation. There are 40 million variations of operative type and the NHS elective waiting list may reach more than 10 million. A coronavirus second wave may cause further delays and expansion of the waiting list. This blog from hub topic lead Richard Jones describes a proven approach to prioritising the waiting list built around individualised risk-adjustment for each patient and evolved from the core POSSUM methodology that is widely used for individual risk assessment pre-operatively. A significant backlog of elective surgical cases has built up during the COVID-19 crisis. The freeze on elective surgery has produced a waiting list that may take years to clear. In the US, the CDC has issued guidelines that "facilities should establish a prioritization policy committee consisting of surgery, anesthesia and nursing leadership to develop a prioritization strategy appropriate to the immediate patient needs". According to the CDC, this committee should work around 'objective priority scoring'. The MeNTS (Medically-Necessary, Time-Sensitive Procedures) instrument is a clever attempt to deliver this scoring, responding to availability of resources and the situation around COVID-19. However, the key challenge is that that the list needs to be prioritised in a way that reflects patient needs and ensures their safety. This is not something that MeNTS can deliver. It also is built around COVID-19 related limitations on resources and this will vary in significance depending on the hospital location and where it is in the journey out of lockdown. The risks of mortality and complications for a patient are a complex combination of the severity of the procedure and the physiological variables of the patient. As an example, a 55-year-old undergoing a radical laproscopic prostatectomy has a risk of mortality of 1.6%. However, if the patient has low blood pressure, that risk triples. If the patient also has low sodium then the risk is 10 times higher [C2-Ai insights]. The spectrum of different operations and key physiological variables creates at least 40 million potential combinations and hence risk. This is hard to manage with one patient but trying to prioritse a group of 5, 10, 100, 1,000 or even 10,000 becomes unmanageable. New patients will be joining the list while others leave following their procedures and so triage of the list will not be a one-off event. The list will need to be populated and triaged intelligently and in a consistent way repeatedly at least until there is a return to ‘normality’. There is evidence that some trusts are attempting to build their own systems for prioritisation. This may be possible around matching operative type and resource availability but the efficiency of these systems overall should be a concern. Best intentions are fine but, when reviewed later, the ability to correctly prioritise patients to minimise harm and mortality is likely to be limited if not flawed. C2-Ai’s COMPASS Surgical List Triage system is an example of a system that can support evidence-based triage and individualised risk assessment of patients, while supporting the objectives of the CDC. It supports clinical decision making across all phases from crisis back to steady state. It has been developed by the creator of the POSSUM system and is built around the world’s largest patient data set (140 million records from 46 countries) through the support of NHS Digital. The underlying algorithms are constantly refined against new and existing data sets to ensure relevance and accuracy. The Surgical List Triage tool combines the mortality and complication risks from the different patients to derive the prioritisation. The system carries out bulk assessments using individualised risk assessments for each patient. These reflect the operative type and their physiology to calculate the risk of mortality and complications, as well as providing a detailed breakdown of potential complications with percentage probability with a simple click. This system also suggests patients that should be reviewed for potential optimisation before any procedure. The physician can click on the link to see the detailed risks for the patient to support their decision making. The system can be used regularly to maintain the logic and integrity of the elective surgical list. This is superior to the potentially fragmented approach where parts of the list are manually considered in isolation as this cannot support effective optimisation of the whole list and the absence of any supporting evidence means the triage will vary enormously. COMPASS SLT is an evidence-based approach that supports optimal ordering of the list and clinical decision making that reduces avoidable harm and mortality. This in turn reduces variation, and cost while freeing bed capacity and also allowing the list to be tackled more quickly. When a patient comes in for the operation, an individual risk-assessment can be done using the COMPASS Pre-Operative Risk Assessment app. This provides a final check on whether the patient’s condition would justify optimising their condition before their procedure. However, it also details the most likely post-procedural complications individualised for the patient and their condition. That allows the treatment pathway to be tailored to that patient as well as recruiting the patient into their own recovery. For example, knowing that chest infection is the highest risk for a patient supports a conversation with them to stress the need for them to get up and about on the day of the operation. As an aside, the risk of mortality and complications can also be used as a strong element in showing informed consent has been obtained from the patient. In combination, these tools can provide a platform to support effective and ongoing triage of the list while reducing harm and unnecessary costs. The systems are currently in use in 12 trusts in the NHS. How are you prioritising waiting lists? We'd be interested to hear and share how you and your trust are dealing with the backlog. Opinions expressed in blogs and other content are those of the author. Patient Safety Learning welcomes sharing content and opinions that promotes safer patient care and for the reduction of avoidable harm. The views expressed on the hub however do not necessarily represent Patient Safety Learning's views or values. References to a specific product or service does not imply a recommendation or endorsement.