Follow-up of indeterminate pulmonary nodules: Points of failure and AI-powered interventions

Millions of non-screening chest CTs are performed yearly for various medical investigations, such as into the causes of cough, shortness of breath, or chest pains. Frequently, these scans will (also) reveal subtle, indeterminate pulmonary nodules (IPNs). And while the majority will be benign, some will be cancerous, a life-changing outcome for hundreds of thousands of people.

Yet, there is a significant challenge in following up on these nodules. Nearly half to most patients with possible pulmonary nodules do not receive proper follow-up, delaying diagnosis or resulting in unnecessary procedures. Where do things go wrong, and how can we do better?

In this article, we look at lung nodule follow-up in routine practice as complementary to targeted lung cancer screening. We then consider potential points of failure in the current follow-up workflow and propose improvements.

Complementary initiatives in lung nodule care

The encouraging results of lung cancer screening trials such as NELSON and NLST are gradually reflected in real-world implementations. England’s Targeted Lung Health Checks, notably, ‘turned a huge corner‘ this year. Since their launch in 2018, 2,000 people have received a lung cancer diagnosis, 76% at an early, potentially curable stage. Backed by a positive recommendation from the EU, more European countries are expected to follow course and launch regional or nationwide initiatives.

Nonetheless, screening alone cannot fully alleviate the burden of the world’s deadliest cancer, mainly because of the programmes’ targeted nature. Only eligible individuals – older and with a smoking history – are offered a precautionary low-dose CT scan of their lungs. At the same time, the lung cancer risk profile of persons with incidentally detected lung nodules differs from that of screening-eligible persons. For example, 10 to 20% of people developing lung cancer have never smoked.

Opportunistic screening in routine clinical practice should, therefore, complement targeted screening. Effective nodule management becomes crucial to diagnosing and managing non-malignant cases as effectively as possible. This is emphasised in clinical recommendations such as the UK’s lung cancer GIRFT report (Getting It Right the First Time).

There are, however, multiple challenges in the lung nodule pathway: the need for robust governance, proper pathway coordination, clarity regarding overall responsibility for the patient, and IT infrastructure to enable virtual clinic follow-up, among others identified in the GIRFT report.

How do these points manifest in a hospital’s workflow and a patient’s journey? Let’s have a closer look.

Points of failure in IPN follow-up

What are “the known unknowns”? Jonathan Rodrigues, Consultant Radiologist at the Royal United Hospitals Bath NHS Foundation Trust, gave a memorable talk with this title during a British Institute of Radiology (BIR) lung cancer imaging study day. In his research, he identified the following potential points of failure in the detection, reporting, and structured follow-up of incidentally detected pulmonary nodules:

1. Suspicious lung nodules on chest CT scans may be missed by the radiologist.

The radiology workforce is under a lot of pressure due to severe understaffing, as confirmed in the latest Royal College of Radiologists census. One of the consequences of the shortage is work overload, which has been linked to diagnostic errors in CT exams.

IPN detection is particularly error-prone. Nodules might be easy to miss due to their size or ‘blind spot’ location. Or, they might not get the clinician’s full attention because identifying them is not the primary purpose of the examination, especially if the scan is performed in an emergency scenario.

2. No actionable recommendation for detected nodules.

Secondly, even if detected, IPNs typically do not get a lot of focus because, as mentioned, most of them will prove to be benign. Health services are already stretched and might avoid the additional burden of actioning a finding that is likely clinically insignificant.

3. The agreed action is not confirmed by the physicians or multidisciplinary team.

Thirdly, including a pulmonary nodule in the impression section of a radiology report does not ensure follow-up. Radiologists and pulmonologists we have spoken to have noticed that the radiologists’ recommendations may be overlooked by the referring physician. It is particularly true if the latter’s primary goal is, for instance, treating the patient for trauma or acute diseases. There is, moreover, no reminder of the action needed at a later stage. In other cases, treating physicians might decide not to refer the patient to the pulmonologist.

Additionally, multidisciplinary team meetings (MDMs) are often busy with complex cases. Nodule patients might be pushed toward the bottom of the agenda or missed altogether.

4. The report actions are not scheduled.

All the above actions checked, we might still be facing one of the main points of failure is IPN follow-up: the lack of communication between the relevant actors in the pathway, from the original radiology reading to the decision maker. There is a level of uncertainty concerning the different responsibilities of follow-up actions for IPNs. Scheduling misses are a reflection of this miscommunication.

5. The patient does not receive timely follow-up.

A consequence of the previous points is delayed patient care for a possibly threatening finding.

The image below offers a summary of these potential moments of failure. We acknowledge that it is not a one-size-fits-all model. The highlighted points vary across practices and countries that use different models of care, different quality standards, or have other clinical capacities and experience.

Improvements to the standard of care

No healthcare practitioner wants patients to fall between one of the cracks in a diagnostic pathway. To make sure that we identify and adequately follow up on every potentially harmful pulmonary nodule, we propose four mitigation strategies:

AI-aided detection

Artificial intelligence (AI) is well-equipped to address the first potential point of failure in the lung nodule pathway: detection. Reporting chest CT scans for lung nodules, radiologists search for millimetric lesions with the naked eye, count, segment, and measure them semi-manually. These tedious tasks are exactly where an AI system can take some of the burden off. For example, Veye Lung Nodules, our automated lung nodule management solution, has been proven to increase the radiologist’s sensitivity.

It’s worth mentioning that Prof. Dr. Marie-Pierre Revel, a reputable thoracic imaging specialist, considers detection error reduction as the AI’s main benefit. She recently emphasised this argument in an interview from ECR 2023.

Active IPN monitoring

A possible intervention for points two to five is a human-filled care coordination role, offered in some healthcare systems, like the Netherlands. Yet, considering workforce shortages and available budgets, technology should also be considered as a solution to the unmet needs. An IT system could be implemented within the existing clinical workflow to track possible lung cancer patients, with features such as:

• Pre-filled lung nodule information based on the AI’s output, including a follow-up recommendation following the British Thoracic Society (BTS) guidelines, which offer the gold standard for lung nodule pathway. Using the same output and guidelines comes with the added benefit of improved adherence and consistency across patients;
• Automatically created follow-up lists;
• Flagging the patient cases about which the radiologist is confident that action is required, e.g., referral to a pulmonologist, overdue appointments, or waiting for a scan.

We have been developing and testing a lung nodule follow-up software, with promising results. An audit conducted at the Royal United Hospitals Bath NHS Foundation Trust consisted of a review of all relevant chest CT scans from a single week. Around 15% of all nodule cases received a questionable clinical recommendation. Extrapolated over the entire year, it amounts to 1,200 patient follow-up decisions that could be improved – in a single hospital.

Many of these decisions would have been improved through the use of the follow-up software with automated patient management recommendations. This has the potential to bring two benefits:

• avoid several unnecessary surveillance scan recalls, which expose patients to avoidable radiation; and
• fast-track others whose nodules needed immediate attention, such as a tissue biopsy or multi-disciplinary review.

The results of the audit and software validation study will be published soon, along with a cost analysis of a virtual nodule team using a thoracic radiologist’s time in this manner.

Nodule clinics

In nodule clinics, a multidisciplinary team evaluates the pulmonary nodules detected in routine practice. They are a practical approach to organising a streamlined patient follow-up, so no early-stage lung cancer is missed.

Together with AstraZeneca, we aim to implement nodule clinics in several hospitals and establish them as the standard of care. In this project, we facilitate communication between radiology and pulmonology by enabling the radiologist to indicate which patients require follow-up and constructing a comprehensive list of patients for the pulmonologist to consider including in the nodule clinic. This approach reduces unnecessary actions and time spent discussing nodules with colleagues from different specialties.

Interestingly, we have noticed that, once a patient has been seen by a pulmonologist, they are unlikely to be referred to the nodule clinic for a detected lung nodule. This makes it difficult to include the total population of patients in one clinic.

Lung nodule malignancy scoring

Beyond detection and measurements, AI can play a role in establishing the nature of detected lung nodules and, subsequently, support the early and accurate diagnosis of lung cancer and the reduction of unnecessary procedures.

In a study published in Nature, scientists affiliated with Google Health presented a highly accurate model for malignancy classification. Our team is working to develop, validate, and bring this model to the market to support early and accurate lung cancer diagnosis.

The challenge of workflow variations

One hurdle to the implementation of the above strategies is one that applies to AI in real-world healthcare in Europe in general: the differences across countries. Organisational structures, data requirements, as well as attitudes toward emerging technologies vary significantly. For an improved IPN process, this challenge manifests in the clinical workflows.

There are significant differences in the route a detected lung nodule would take in different geographies. For instance, in the UK’s National Health Service (NHS), the next action typically involves sending the case, including the radiologist’s recommendation, directly to an expert thoracic consultant to determine the next action. In Germany and the Netherlands, putting a case forward on the care pathway is not possible in one step. Instead, the case is first returned to the referring physician. In Spain, nodules are managed by different departments, such as internal medicine, making it difficult to centralise the process.

Therefore, incorporating an AI-driven solution for IPN follow-up requires a tailored rather than a plug-and-play approach.

Problems and solutions

Improving the follow-up of indeterminate pulmonary nodules is a complex undertaking. Yet we have good reasons to see it through.

Nine in ten people survive when diagnosed with lung cancer at stage one compared to only two in ten people at stage four. Every suspicious lung nodule identified in routine clinical practice is an opportunity to deliver an early, possible life-saving diagnosis.

The potential points of failure in following up on IPNs are known. They may be overlooked nodules, omission of actionable recommendations, lack of communication around the report actions, or scheduling misses. At the same time, technological solutions to mitigate these points are emerging. Two examples are AI for automated lung nodule management and IPN monitoring software.

As the burden of cancer is drastically increasing in our society, we must match the problems and the solutions and make sure no lung cancer patient is left behind.