Patients’ and dermatologists’ preferences in artificial intelligence–driven skin cancer diagnostics: A prospective multicentric survey study
Sarah Haggenmüller, Roman C. Maron, Achim Hekler, Eva Krieghoff‐Henning, Jochen Utikal, Maria Rita Gaiser, Verena Müller, Sascha Fabian, Friedegund Meier, Sarah Hobelsberger, Frank Friedrich Gellrich, Mildred Sergon, Axel Hauschild, Michael Weichenthal, Lars E. French, Lucie Heinzerling, Justin Gabriel Schlager, Kamran Ghoreschi, Max Schlaak, Franz J. Hilke, Gabriela Poch, Sören Korsing, Carola Berking, Markus V. Heppt, Michael Erdmann, Sebastian Haferkamp, Konstantin Drexler, Dirk Schadendorf, Wiebke Sondermann, Matthias Goebeler, Bastian Schilling, Jakob N. Kather, Stefan Fröhling, Katharina Kaminski, Astrid Doppler, Tabea-Clara Bucher, Titus J. Brinker, Carina Nogueira Garcia, Janis Thamm, Sandra Schuh, Julia Welzel, Lara Valeska Maul-Duwendag, Paul Georg, Laurence Feldmeyer, Falk G. Bechara, Julian Steininger, Sophia Lehr, Ricarda Rauschenberg, Anna-Lisa Eberle, Sören Hartmann, Helena Hasler, Sophia Bender-Säbelkampf, Jürgen Bauerschmitz, Matthias D. Kaufmann, Cornelia Erfurt‐Berge, Wiebke K. Peitsch, Ulrike Wehkamp, Marion Jost, Cindy Franklin, Julia Holzgruber, Maximilian Haist, Mario Giulini, Sebastian A. Wohlfeil, Valentina Faihs, Elke Sattler, Suzan Stürmer, Sebastian Krammer, Benjamin Kendziora, Semra Larissa Akcetin, Mohammed Mitwalli, Pinar Avci, Zeno Fiocco, Daniela Hartmann, Manuel P. Pereira, Alexander Thiem, Valentin Aebischer, Stephan Forchhammer, Isabel Wolff
Abstract
To the Editor: Artificial intelligence (AI) has shown promise for improving diagnostics of skin cancer by matching or surpassing experienced clinicians.1 However, the successful clinical application depends on acceptance by patients and dermatologists. In this prospective multicentric survey study with a response rate of 63%, we therefore investigate the criteria required for patients and dermatologists to accept AI-systems and assess their importance on patients’ and dermatologists’ decision-making when considering the use of such systems. To this end, we perform an adaptive choice-based conjoint analysis and analyze it using hierarchical Bayes estimation.2 By employing an adaptive choice-based conjoint analysis, we investigate multiple influencing AI-features simultaneously (see Table I) whilst accounting for possible trade-offs (see Fig 1). For details on questionnaire development, participant recruitment, and statistical analysis, see Supplementary Methods, available via Mendeley at https://data.mendeley.com/datasets/2chcwnhpwj/1.