At a time when artificial intelligence is predicted to be a powerful tool for early detection of lung cancer (BC), researchers are questioning the belief that the growing supply of additional information is really always useful and asking, on the contrary, if using less, but more relevant, data would not be more useful for medical decisions? This reasoning is based on heuristics which make it possible to develop simple strategies deliberately ignoring certain information.
LE Kearney et al. (1) present CB screening strategy (SD) models compared to those proposed by the American USPSTF working group (US Preventive Services Task Force). Data obtained by the USPSTF for BC screening by low-dose chest computed tomography (LTFD) in 71,978 subjects were reanalyzed according to a “rapid and modest” heuristic (HRM) establishing simple SD algorithms (fast-and-frugal tree algorithms).
The USPSTF organized the BC SD according to numerous demographic and anamnestic criteria of smoking history. This working group also established that the best criterion for the effectiveness of an SD is the measurement of the gain in additional days of life, established at 16.2 days and defining the so-called “significant” benefit threshold of the SD, under reserve of 3 annual screening examinations and a 5-year follow-up. Despite all these efforts, the results obtained are quite slim and not without limitations (2). The HRM proposed by LE Kearney et al. (1) is a theoretical projection of the USPSTF data: from the 71,978 subjects in the study, 69,395 were retained for the analysis and the calculations of sensitivity and specificity of the 2 protocols covered 31,975 cases. Two trees were established shown below.
The first tree concerns the population whose smoking is > 40 P/A. The comparative sensitivities and specificities of HRM and USPSTF are 91%, respectively. vs 78% (pth decision tree is more specifically devoted to subjects aged 50 to 80 years. Two elements allow this tree to gain in sensitivity and specificity with regard to the USPSTF: smoking > 20 P/A and permanent weaning 15 years old. It makes it possible to include in the SD groups excluded by the USPSTF criteria: subjects having smoked over 40 years but smoking quantification 15 years, subjects of African-American origin often being in this case. The comparative sensitivity of HRM and USPSTF is then 83% versus 56% (p
The authors conclude that screening strategies using simple criteria can better identify candidates who will benefit the most from lung cancer screening, while respecting the ethical principle: “equal risk, equal care”. Furthermore, this could substantially reduce the disparities in the organization of screening, neglecting certain minorities.
Health
Related News :