Early detection and lower costs with AI-driven lung cancer screening
Lung cancer is one of the most common cancer diseases and the deadliest form of cancer in terms of number of deaths. A prerequisite for implementing a successful screening program is sufficient capacity in all stages of the care process. A particular concern is the availability of radiologists.
Vision Zero Cancer is a Vinnova-funded collaboration aimed at creating an innovation environment to establish conditions for transforming cancer care. The innovation environment works towards the vision of transforming cancer from a deadly to a curable or chronic disease, where innovative solutions and approaches in several parts of society are required to achieve the vision. Lung cancer is an initial focus area for Vision Zero Cancer.
On behalf of Vision Zero Cancer and AstraZeneca, the Swedish Institute for Health Economics (IHE) has produced an analysis of the conditions for national screening, focusing on radiology, and through a cost-consequence analysis highlighted the effects of implementing various screening scenarios. A future outlook on potential lung cancer screening may include new methods involving AI. The scenario development should be seen as a hypothetical and exploratory development of a screening program for lung cancer.
Summary
Lung cancer is one of the most common cancers; it is also the deadliest form of cancer in terms of number of deaths. During the period 2014–2018, lung cancer was the 4th most common form of cancer in both men and women. Median age at diagnosis is 69 years and estimated relative 5-year survival is approximately 20%.
In Sweden, there is currently no national screening program for lung cancer. Previous studies from the Netherlands and the USA have shown that targeted screening for lung cancer reduces mortality related to the disease. Based on this research indicating advantageous results of screening, together with the ambition of Swedish healthcare to offer cancer patients equal and timely treatment, pilot studies of lung cancer screening have been initiated in some regions and their respective regional cancer centers.
A prerequisite for successful implementation of a screening program is sufficient capacity in all parts of the health care process. A particular concern is the availability of radiologists for assessment of results from computed tomography, as radiologists are a scarce resource. The lack of radiologists and radiology nurses, which has been a problem for many years, create limitations in healthcare in general and cancer care in particular. Seen from a larger societal perspective, the lack of qualified personnel is in line with an overall challenge in finding the right competencies that municipalities and regions are facing already and will continue facing in the coming ten years. All other things being equal to the current situation, the welfare sector including health care, needs to employ a total of nearly 400,000 people by 2031 to meet the demographic development in Sweden. In view of the challenges the Swedish healthcare system is facing, together with the aim to provide equal care for the entire population, it is of interest to evaluate alternative ways of health care procedures.
The possibility of using artificial intelligence (AI) to supplement and support radiology is under discussion and investigated in some lung cancer studies. Parallel to the development in other areas, for example breast cancer screening, it is of interest to investigate a possible screening process for lung cancer including how it could theoretically develop in the future.
The purpose of the study is to analyse the conditions for lung cancer screening with a focus on radiology as well as investigating the effect of different screening scenarios on radiology capacity. In Swedish healthcare, AI is being evaluated as a supporting method in different areas including diagnostics. Future possibilities for lung cancer screening could include new methods such as AI. The scenarios for screening in this study are intended to illustrate a possible and gradual development towards a screening process where all Swedish healthcare regions use AI through a centralized national structure. The scenario development should be seen as a hypothetical and exploratory development of a screening program for lung cancer. At present, more research on AI is needed, including how AI could complement radiology and other areas of the Swedish healthcare system.
The first part of the study aims to identify resources in terms of radiologist and radiology nurses, followed by an analysis of the consequences on radiology of introducing lung cancer screening (Scenario 1, conventional lung cancer screening). This is then compared with scenarios where AI is used to different degrees for lung cancer screening (Scenario 2 and scenario 3). In this study conventional screening refers to a screening process where the work is carried out by healthcare personnel with today's medical technology. The scenarios differ in terms of methods used for reviewing radiological images. In scenario 2, the assumptions are a screening process carried out mainly by the radiologist but with some support from AI. A screening process like scenario 3, where AI is more integrated n the radiology process, is a hypothetical long-term goal of what the screening process for lung cancer could look like in the future. Tasks of larger volume, such as image reviews, could in theory be handled through a national and central AI structure. Such a structure intends to use radiology competence in a new way in parallel with compensating, to some extent, for the lack of radiologists in Sweden.
To estimate the relevant population for screening, the Swedish planning group for lung cancer’s (Svenska planeringsgruppen för Lungcancer, SLUSG) proposal for screening criteria is used. Conversations have been held with radiologists for matters requiring clinical knowledge and practices, for describing the development of radiology and visions about AI in the screening process.
In this study, we include the cost of the radiologist's time as salary excluding overhead costs. The study does not consider other costs such as healthcare personnel, equipment, or other peripheral costs related to a screening program. Likewise, this study does not include any estimate of the development costs of AI or assessment of AI as a tool in healthcare.
For each screening scenario, from 1 to 3, there is a reduction in the cost of the radiologist resource in parallel with a gradual introduction of AI in the radiology process. In Scenario 3, the costs for the radiologist are halved compared to Scenario 2 and approximately one fifth compared to Scenario 1.
Scenario 3 provides the largest cost reduction for the radiologist resource. It is important to bear in mind that the regional conditions such as competence, number of radiologists, infrastructure etc. also play a major role in the conditions for introducing and developing a screening program.
In today's situation, it is not realistic to carry out screening programs where the entire target population is screened without noticing significant regional differences. Even in the regions with the best conditions, given the radiological situation, it would be very challenging. Although previous analyses have shown that lung cancer screening is cost-effective, there are other challenges. A gradual introduction of a screening program or other frequency of screening intervals of the population could be considered. In addition to this, there must be a capacity to take care of secondary findings that are not lung cancer and arise as a result of a screening program.
A national AI structure for image review would reduce resource requirements for radiologists and contribute to more even capacity between regions. This is, however, not realistic in the near future as more research on AI is needed before an eventual introduction. Until then, a screening program would need to be targeted at a subset of the population according to an appropriate screening interval
The report is in Swedish!