Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

ePoster Display

1859P - Using deep learning with demographic and laboratory values from baseline to cycle 2 to predict subsequent renal and hepatic function

Date

16 Sep 2021

Session

ePoster Display

Topics

Management of Systemic Therapy Toxicities;  Supportive Care and Symptom Management

Tumour Site

Breast Cancer;  Colon and Rectal Cancer

Presenters

Matthew Watson

Citation

Annals of Oncology (2021) 32 (suppl_5): S1237-S1256. 10.1016/annonc/annonc701

Authors

M.S. Watson1, P. Chambers2, K. Shiu3, J.A. Bridgewater4, M. Desai2, R. Roylance2, A. Tailor2, S. Masento2, M. Forster5, N. Al Moubayed1

Author affiliations

  • 1 Department Of Computer Science, Durham University, DH1 3LE - Durham/GB
  • 2 Cancer Services, University College London Hospital NHS Foundation Trust, WC1N 1AX - London/GB
  • 3 Gastrointestinal Oncology Service, University College London Hospital NHS Foundation Trust, WC1N 1AX - London/GB
  • 4 Research Department Of Oncology, UCL Cancer Institute, WC1E 6DD - London/GB
  • 5 Department Of Oncology, University College London Hospitals NHS Foundation Trust, NW1 2PG - London/GB

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 1859P

Background

Renal and hepatic function are among the main determinants of chemotherapy drug exposure, meaning possible altered pharmacokinetic profiles in those with impaired organ function, necessitating dose adjustments. Practice is therefore to assess function at each treatment cycle; however, this can cause delays in patients receiving treatment. We aimed to determine if demographic and laboratory values from the first two treatment cycles could predict changes in subsequent renal and hepatic function, for subsequent treatment dosing.

Methods

Data from one large hospital was extracted for age, comorbidity, cancer type, treatment, and laboratory values for each treatment cycle received including creatinine, bilirubin, neutrophils and platelets. Using deep learning, we developed a Multilayer Perceptron regression model to predict creatinine and bilirubin values; these predictions are then used to classify patients into groupings using common-terminology-criteria-for-adverse-events. The F1 score was calculated to determine a balanced measure of accuracy of the model.

Results

Of 962 eligible patients, 334 were male (35%) and 626 female (65%), with a median age of 56 years. Tumour sites were breast cancer (n=323, 32.4%), colorectal cancer (n=362, 36.2%), and diffuse-large-b-cell lymphoma (n=277, 27.5%). All treatments were the first definitive treatment. 90 (9.4%) patients saw grade changes in creatinine, and 72 (7.5%) in bilirubin. The developed prediction models using laboratory values from baseline to cycle 2 showed good performance; on creatinine, we attained an F1 score of 0.8 and precision of 1, and on bilirubin an F1 of 0.85 and precision of 0.92. We estimate that around 70% of patients could be saved from having creatinine and bilirubin tests from cycle 3 onwards.

Conclusions

The developed models will be validated and used as an add on to e-prescribing systems, enabling accurate stratification of patients that need more intense renal and hepatic assessments. We estimate we will save 70% of assessments for patients, with the remaining 30% identified as needing further intensive monitoring. This would lead to fewer delays and improved patient experience.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

UCLH NHS Foundation Trust.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.