Prediction of Decline in Activities of Daily Living Through Deep Artificial Neural Networks and Domain Adaptation
- 223 Downloads
In order to improve information available at the clinical level and to better focus resources for preventive interventions, it is paramount to estimate the general exposure to risk of adverse health events, commonly referred as frailty. This study compares the performance of shallow and deep multilayer perceptrons (sMLP and dMLP), and of long short-term memories (LSTM), on the prediction of a subject decline in activities of daily living, with and without a previous autoencoder based domain adaptation from an external dataset. Samples originates from two large epidemiological datasets: the English Longitudinal Study of Ageing (ELSA) and The Irish Longitudinal Study on Ageing, with 107879 and 15710 eligible samples, respectively. Deep networks performed better than shallow ones, while dMLP and LSTM performance were similar. Domain adaptation improved predictive ability in all comparisons. On the bigger ELSA dataset, sMLP attains a Brier score of 0.32 without domain adaptation, and 0.15 with domain adaptation, while dMLP attains 0.20 and 0.11, respectively. Thus, experimental results support the use of deep architectures in the prediction of functional decline, and of domain adaptation when data from another similar domain is available. These results may help improving the state of the art in predictive models for clinical practice and population screening.
KeywordsArtificial neural networks Deep learning Domain adaptation Frailty Risk assessment Transfer learning
The data relative to ELSA were made available through the United Kingdom Data Archive - www.data-archive.ac.uk. ELSA was developed by a team of researchers based at the NatCen Social Research, University College London and the Institute for Fiscal Studies. The data were collected by NatCen Social Research. The funding is provided by the National Institute of Aging in the United States, and a consortium of United Kingdom government departments coordinated by the Office for National Statistics.
TILDA is an interinstitutional initiative led by Trinity College Dublin. TILDA data have been co-funded by the Government of Ireland through the Office of the Minister for Health and Children, by Atlantic Philanthropies, and by Irish Life; have been collected under the Statistics Act, 1993, of the Central Statistics Office. The project has been designed and implemented by the TILDA study team, Department of Health and Children. Copyright and all other intellectual property rights relating to the data are vested in TILDA. Ethical approval for each wave of data collection is granted by the Trinity College Research Ethics Committee. TILDA data is accessible for free from the following sites: Irish Social Science Data Archive at University College Dublin http://www.ucd.ie/issda/data/tilda/; Interuniversity Consortium for Political and Social Research at the University of Michigan (http://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/34315).
The original data creators, depositors or copyright holders, the funders of the data collections and the archives of the datasets bear no responsibility for their further analysis or interpretation presented here.
Conflict of Interest
All authors declare no competing interests and to be aware of the submission of this manuscript.
- 1.Bengio, Y.: Deep learning of representations for unsupervised and transfer learning. In: JMLR Workshop Conference Proceedings, vol. 7, pp. 1–20 (2011). https://doi.org/10.1109/IJCNN.2011.6033302
- 2.Bengio, Y.: Practical recommendations for gradient-based training of deep architectures. CoRR abs/1206.5533 (2012). http://arxiv.org/abs/1206.5533
- 4.Bengio, Y., Boulanger-Lewandowski, N., Pascanu, R.: Advances in optimizing recurrent networks. In: 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 8624–8628. IEEE (2013)Google Scholar
- 8.Brier, G.W.: Verification of forecasts expressed in terms of probability. Mon. Weather Rev. 78, 1 (1950). https://doi.org/10.1175/1520-0493(1950)078<0001:VOFEIT>2.0.CO;2CrossRefGoogle Scholar
- 10.Buz, J., Cortés-Rodríguez, M.: Measurement of the severity of disability in community-dwelling adults and older adults: interval-level measures for accurate comparisons in large survey data sets. BMJ Open 6(9), e011842 (2016). https://doi.org/10.1136/bmjopen-2016-011842, https://bmjopen.bmj.com/content/6/9/e011842CrossRefGoogle Scholar
- 13.Csurka, G.: Domain adaptation for visual applications: a comprehensive survey, pp. 1–46. CoRR abs/1702.05374 (2017). http://arxiv.org/abs/1702.05374
- 18.Donati, L.: Domain adaptation through deep neural networks for health informatics (2017)Google Scholar
- 20.Fongo, D.: Previsione del declino funzionale tramite l’utilizzo di reti neurali ricorrenti (2017)Google Scholar
- 21.Glorot, X., Bengio, Y.: Understanding the difficulty of training deep feedforward neural networks. In: Teh, Y.W., Titterington, M. (eds.) Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics. Proceedings of Machine Learning Research, vol. 9, pp. 249–256. PMLR, Chia Laguna Resort, Sardinia, Italy, 13–15 May 2010. http://proceedings.mlr.press/v9/glorot10a.html
- 22.Glorot, X., Bordes, A., Bengio, Y.: Domain adaptation for large-scale sentiment classification: a deep learning approach. In: Proceedings of the 28th International Conference on Machine Learning, no. 1, pp. 513–520 (2011). http://www.icml-2011.org/papers/342_icmlpaper.pdf
- 27.Banks, J., Batty, G.D., Nazroo, J., Steptoe, A.: The dynamics of ageing: evidence from the English Longitudinal Study of Ageing 2002–15 (Wave 7). The Institute for Fiscal Studies (2016)Google Scholar
- 28.Kenny, R.A.: The Irish longitudinal study on ageing (TILDA) 2009–2011 (2014). https://doi.org/10.3886/ICPSR34315.v1
- 29.Kenny, R.A., et al.: The design of the Irish longitudinal study on ageing. Lifelong Learn. (2010)Google Scholar
- 30.Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. CoRR abs/1412.6980 (2014). http://arxiv.org/abs/1412.6980
- 37.Lipton, Z.C., Kale, D.C., Elkan, C., Wetzell, R.: Learning to diagnose with LSTM recurrent neural networks. In: ICLR, pp. 1–18 (2015). http://arxiv.org/abs/1511.03677
- 44.Purushotham, S., Carvalho, W., Nilanon, T., Liu, Y.: Variational adversarial deep domain adaptation for health care time series analysis. In: 29th Conference on Neural Information Processing System (NIPS) (2016). https://wcarvalho.github.io/files/nips_2016/VADA_main.pdf
- 47.Robert, C., Arreto, C.D., Azerad, J., Gaudy, J.F.: Bibliometric overview of the utilization of artificial neural networks in medicine and biology. Scientometrics 59(1), 117–130 (2004). https://doi.org/10.1023/B:SCIE.0000013302.59845.34CrossRefGoogle Scholar
- 48.Song, X., Mitnitski, A., Cox, J., Rockwood, K.: Comparison of machine learning techniques with classical statistical models in predicting health outcomes. Medinfo 11, 736–740 (2004)Google Scholar
- 53.Vincent, P., Larochelle, H., Bengio, Y., Manzagol, P.A.: Extracting and composing robust features with denoising autoencoders. In: Proceedings of 25th International Conference on Machine Learning, ICML 2008, pp. 1096–1103 (2008). https://doi.org/10.1145/1390156.1390294, http://portal.acm.org/citation.cfm?doid=1390156.1390294
- 54.Vincent, P., Larochelle, H., Bengio, Y., Manzagol, P.A.: Extracting and composing robust features with denoising autoencoders. In: Proceedings of the 25th International Conference on Machine Learning, ICML 2008, pp. 1096–1103. ACM, New York (2008). https://doi.org/10.1145/1390156.1390294