Background and Aims: Detecting NASH remains challenging, while at-risk NASH (steatohepatitis and F≥ 2) tends to progress and is of interest for drug development and clinical application. We developed prediction models by supervised machine learning techniques, with clinical data and biomarkers to stage and grade patients with NAFLD. Approach and Results: Learning data were collected in the Liver Investigation: Testing Marker Utility in Steatohepatitis metacohort (966 biopsy-proven NAFLD adults), staged and graded according to NASH CRN. Conditions of interest were the clinical trial definition of NASH (NAS ≥ 4;53%), at-risk NASH (NASH with F ≥ 2;35%), significant (F ≥ 2;47%), and advanced fibrosis (F ≥ 3;28%). Thirty-five predictors were included. Missing data were handled by multiple imputations. Data were randomly split into training/validation (75/25) sets. A gradient boosting machine was applied to develop 2 models for each condition: clinical versus extended (clinical and biomarkers). Two variants of the NASH and at-risk NASH models were constructed: direct and composite models. Clinical gradient boosting machine models for steatosis/inflammation/ballooning had AUCs of 0.94/0.79/0.72. There were no improvements when biomarkers were included. The direct NASH model produced AUCs (clinical/extended) of 0.61/0.65. The composite NASH model performed significantly better (0.71) for both variants. The composite at-risk NASH model had an AUC of 0.83 (clinical and extended), an improvement over the direct model. Significant fibrosis models had AUCs (clinical/extended) of 0.76/0.78. The extended advanced fibrosis model (0.86) performed significantly better than the clinical version (0.82). Conclusions: Detection of NASH and at-risk NASH can be improved by constructing independent machine learning models for each component, using only clinical predictors. Adding biomarkers only improved the accuracy of fibrosis.
Machine learning algorithm improves the detection of NASH (NAS-based) and at-risk NASH: A development and validation study / J. Lee, M. Westphal, Y. Vali, J. Boursier, S. Petta, R. Ostroff, L. Alexander, Y. Chen, C. Fournier, A. Geier, S. Francque, K. Wonders, D. Tiniakos, P. Bedossa, M. Allison, G. Papatheodoridis, H. Cortez-Pinto, R. Pais, J. Dufour, D.J. Leeming, S. Harrison, J. Cobbold, A.G. Holleboom, H. Yki-Järvinen, J. Crespo, M. Ekstedt, G.P. Aithal, E. Bugianesi, M. Romero-Gomez, R. Torstenson, M. Karsdal, C. Yunis, J.M. Schattenberg, D. Schuppan, V. Ratziu, C. Brass, K. Duffin, K. Zwinderman, M. Pavlides, Q.M. Anstee, P.M.1. Bossuyt, Q.M. on behalf of the LITMUS investigators Anstee, A.K. Daly, O. Govaere, S. Cockell, D. Tiniakos, P. Bedossa, A. Burt, F. Oakley, H.J. Cordell, C.P. Day, K. Wonders, P. Missier, M. Mcteer, L. Vale, Y. Oluboyede, M. Breckons, P.M. Bossuyt, H. Zafarmand, Y. Vali, J. Lee, M. Nieuwdorp, A.G. Holleboom, J. Verheij, V. Ratziu, K. Clément, R. Patino-Navarrete, R. Pais, V. Paradis, D. Schuppan, J.M. Schattenberg, R. Surabattula, S. Myneni, B.K. Straub, T. Vidal-Puig, M. Vacca, S. Rodrigues-Cuenca, M. Allison, I. Kamzolas, E. Petsalaki, M. Campbell, C.J. Lelliott, S. Davies, M. Orešič, T. Hyötyläinen, A. Mcglinchey, J.M. Mato, Ó. Millet, J. Dufour, A. Berzigotti, M. Masoodi, M. Pavlides, S. Harrison, S. Neubauer, J. Cobbold, F. Mozes, S. Akhtar, S. Olodo-Atitebi, R. Banerjee, M. Kelly, E. Shumbayawonda, A. Dennis, A. Andersson, I. Wigley, M. Romero-Gómez, E. Gómez-González, J. Ampuero, J. Castell, R. Gallego-Durán, I. Fernández, R. Montero-Vallejo, M. Karsdal, D.G.K. Rasmussen, D.J. Leeming, A. Sinisi, K. Musa, E. Sandt, M. Tonini, E. Bugianesi, C. Rosso, A. Armandi, F. Marra, A. Gastaldelli, G. Svegliati, J. Boursier, S. Francque, L. Vonghia, A. Driessen, M. Ekstedt, S. Kechagias, H. Yki-Järvinen, K. Porthan, J. Arola, S. van Mil, G. Papatheodoridis, H. Cortez-Pinto, C.M.P. Rodrigues, L. Valenti, S. Pelusi, S. Petta, G. Pennisi, L. Miele, A. Geier, C. Trautwein, G.P. Aithal, S. Francis, P. Hockings, M. Schneider, P. Newsome, S. Hübscher, D. Wenn, C. Rosenquist, A. Trylesinski, R. Mayo, C. Alonso, K. Duffin, J.W. Perfield, Y. Chen, C. Yunis, T. Tuthill, M.A. Harrington, M. Miller, Y. Chen, E.J. Mcleod, T. Ross, B. Bernardo, C. Schölch, J. Ertle, R. Younes, A. Oldenburger, R. Ostroff, L. Alexander, H. Biegel, M.S. Kjær, L.M. Harder, P. Davidsen, L.F. Mikkelsen, M. Balp, C. Brass, L. Jennings, M. Martic, J. Löffler, D. Applegate, S. Shankar, R. Torstenson, C. Fournier-Poizat, A. Llorca, M. Kalutkiewicz, K. Pepin, R. Ehman, G. Horan, G. Ho, D. Tai, E. Chng, S.D. Patterson, A. Billin, L. Doward, J. Twiss, P. Thakker, H. Landgren, C. Lackner, A. Gouw, P. Hytiroglou. - In: HEPATOLOGY. - ISSN 1527-3350. - 78:1(2023 Jul), pp. 258-271. [10.1097/HEP.0000000000000364]
Machine learning algorithm improves the detection of NASH (NAS-based) and at-risk NASH: A development and validation study
L. Valenti;
2023
Abstract
Background and Aims: Detecting NASH remains challenging, while at-risk NASH (steatohepatitis and F≥ 2) tends to progress and is of interest for drug development and clinical application. We developed prediction models by supervised machine learning techniques, with clinical data and biomarkers to stage and grade patients with NAFLD. Approach and Results: Learning data were collected in the Liver Investigation: Testing Marker Utility in Steatohepatitis metacohort (966 biopsy-proven NAFLD adults), staged and graded according to NASH CRN. Conditions of interest were the clinical trial definition of NASH (NAS ≥ 4;53%), at-risk NASH (NASH with F ≥ 2;35%), significant (F ≥ 2;47%), and advanced fibrosis (F ≥ 3;28%). Thirty-five predictors were included. Missing data were handled by multiple imputations. Data were randomly split into training/validation (75/25) sets. A gradient boosting machine was applied to develop 2 models for each condition: clinical versus extended (clinical and biomarkers). Two variants of the NASH and at-risk NASH models were constructed: direct and composite models. Clinical gradient boosting machine models for steatosis/inflammation/ballooning had AUCs of 0.94/0.79/0.72. There were no improvements when biomarkers were included. The direct NASH model produced AUCs (clinical/extended) of 0.61/0.65. The composite NASH model performed significantly better (0.71) for both variants. The composite at-risk NASH model had an AUC of 0.83 (clinical and extended), an improvement over the direct model. Significant fibrosis models had AUCs (clinical/extended) of 0.76/0.78. The extended advanced fibrosis model (0.86) performed significantly better than the clinical version (0.82). Conclusions: Detection of NASH and at-risk NASH can be improved by constructing independent machine learning models for each component, using only clinical predictors. Adding biomarkers only improved the accuracy of fibrosis.
| File | Dimensione | Formato | |
|---|---|---|---|
|
machine_learning_algorithm_improves_the_detection.21.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
2.25 MB
Formato
Adobe PDF
|
2.25 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
