Portal de Periódicos da CAPES

Cost of artificial intelligence: Elephant in the room and its cage

2023; Wiley; Volume: 35; Issue: 7 Linguagem: Inglês

10.1111/den.14567

1443-1661

Cesare Hassan, Alessandro Repici, Yuichi Mori,

Gastric Cancer Management and Outcomes

The use of computer-assisted diagnosis (CADe) for screening colonoscopy is now supported by several randomized trials. A substantial decrease in the miss rate of neoplasia coupled with a substantial increase in both of the key quality indicators, namely adenoma detection rate (ADR) and adenomas per colonoscopy, have been successfully and consistently achieved by CADe. This short-term clinical effectiveness is expected to contribute to prevention of colorectal cancer in the long run. No innovation in the field of advanced imaging of colonoscopy has ever been supported by such an impressive amount of evidence. Nevertheless, the CADe tools have not been well implemented outside the tertiary centers. Although more than 10 CADe devices from different manufacturers are available on the global market, its penetration in community-based hospitals is quite discouraging, with possible adoption of less than 1% of all the screening centers. Most of the organized screening programs, endoscopic centers, and participants of the screening programs do not have access to a potential life-saving intervention. Why is this situation happening? There are multiple factors that are converging on an instantaneous spreading of CADe at a similar extent as any successful App on our mobile devices. For instance, these CADe devices are designed to get integrated into clinical practice in a plug-and-play and user-friendly fashion. No training is basically needed before we appreciate the benefits of using CADe. Furthermore, patients are usually enthusiastic to be protected by CADe once they are informed of the substantial risk of error of the endoscopist due to distraction, tiredness, or lack of competence. The answer to the previous question may be quite straightforward. The most probable explanation is the cost of CADe and the lack of the reimbursement from the health-care systems, health insurance bodies, and authorities. We must be honest about the following two facts: (i) CADe is far from being cheap; and (ii) hospitals seldom use medical devices/drugs that are not supported by reimbursement. Research and development of artificial intelligence software takes tremendous effort and time and it is always subject to the need of updating, both of which lead to the high price of the devices. It probably represents one of the most expensive endoscopy technologies when excluding the tower and the scopes. In addition, such cost must be multiplied by the actual number of towers any endoscopic center owns. Hospitals usually seek financial support or reimbursement when investing in such resources; however, there has been no such established reimbursement system for the CADe tools. This is absolutely different from the situation of the CADe tools for mammography, which is generally subject to reimbursement in the US health-care system.1 However, establishment of new reimbursement for a new medical device is a big challenge in any of the health-care systems simply because they are always struggling with financial matters. Therefore, it is quite important to show the cost-effectiveness of the new intervention in various settings. Any time there is a price-sensitive scenario, a meticulous comparison between the benefit and the cost must be looked for. The benefit should not be simplified in a simple financial return, as benefit in health policy is considered as any gain in quality or quantity of life, namely, quality-adjusted life-years. When such benefit is achieved at a reasonably acceptable cost, namely, the willingness-to-pay threshold, the intervention is considered to be worthy of investment. In this issue of Digestive Endoscopy, Sekiguchi et al.2 addressed the cost-effectiveness of CADe when implemented in an organized program based on an invitation to the fecal immunochemical test (FIT) in Japan. By projecting a real-life gradient of ADR observed in a Japanese study on the natural history of colorectal cancer, the authors showed the convenience of CADe for post-FIT+ colonoscopy as the additional cost of the intervention divided by the quality-adjusted life-years saved by the additional CADe-driven detection of colorectal neoplasia was much less than the accepted threshold of cost-effectiveness. The relevance of such a finding is to confirm in an organized setting the favorable cost-effectiveness of CADe already shown in a primary screening colonoscopy setting. When considering that in several European countries, as well as in Japan, millions of people are actively invited to FIT screening, any intervention that is effective and efficient in this setting may be expected to result in a substantial absolute benefit. The timeliness of this finding is that its publication corresponds with two recently published articles showing, for the first time, the validity of the inverse association between ADR and colorectal cancer (CRC) in a FIT setting.3, 4 As the assumption of CADe in this simulation was purely based on the ADR gradient achieved by CADe colonoscopy, the recent validation of ADR-CRC in FIT programs substantiates the results of the model by Sekiguchi et al.2 Does this mean that the elephant in the room is finally in the cage and that we can start to massively invest in CADe technology within an organized screening program? Unfortunately, there are several residual uncertainties that must be addressed before definitively closing the cage. First, there is a hierarchical difference between benefit and costs in this as well as other simulations. Costs are certain and immediate, while benefits are uncertain and far in time, i.e. 20–30 years after the interventions. Thus, it is unclear whether any health systems are ready to wait for such a long time in order to be confident in the value of their investment. Second, there is uncertainty whether an ADR increase driven by CADe is actually able to decrease in the long term the incidence and mortality of CRC. For instance, CADe could be only able to detect diminutive innocent adenomas that would never turn into cancer, without affecting the detection of more relevant advanced neoplasia. The lack of evidence on the efficacy of CADe in increasing the detection of advanced adenomas is much more worrisome in the setting of organized programs, as the main filter effect of FIT is on the advanced adenomas rather than on those nonadvanced. Thus, we urgently need studies showing the long-term efficacy of CADe in preventing incidence and mortality of CRC before claiming the validity of such simulation models. Fortunately, one large-scale study on this topic is ongoing in Europe.5 Third, there are alarming data when passing from a control to a real-life setting. At least two studies showed the failure of CADe in increasing ADR outside the setting of clinical trials,6, 7 questioning the generalizability of the result of a randomized clinical trial. In conclusion, the evidence for a favorable cost-effectiveness profile of CADe in the setting of an organized program is reassuring. However, this should be considered as a prompt for launching new research in this field rather than as conclusive support for its widespread implementation. Authors C.H. and A.R.: Medtronic Corporation (consultancy, equipment on loan), Fujifilm Corporation (consultancy, equipment on loan), and Pentax Corporation (consultancy). Y.M.: Olympus Corporation (speaker fee, consultancy, equipment on loan) and Cybernet Systems Corporation (ownership interest). None.