June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Predicting optimal treatment regimen for patients with neovascular age-related macular degeneration (nAMD) using machine learning
Author Affiliations & Notes
  • Yusuke Kikuchi
    Genentech Inc, South San Francisco, California, United States
    University of California Berkeley Department of Industrial Engineering and Operations Research, Berkeley, California, United States
  • Ales Neubert
    Roche Pharma Research and Early Development Informatics, Basel, Switzerland
  • Jian Dai
    Genentech Inc, South San Francisco, California, United States
  • Carlos Quezada Ruiz
    Genentech Inc, South San Francisco, California, United States
  • Footnotes
    Commercial Relationships   Yusuke Kikuchi Genentech, Inc., Code E (Employment); Ales Neubert F. Hoffmann-La Roche Ltd., Code E (Employment); Jian Dai Genentech, Inc., Code E (Employment); Carlos Quezada Ruiz Genentech, Inc., Code E (Employment)
  • Footnotes
    Support  Yes, Genentech, Inc., South San Francisco, CA, provided support for the study and participated in the study design; conducting the study; and data collection, management, and interpretation
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2986 – F0256. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yusuke Kikuchi, Ales Neubert, Jian Dai, Carlos Quezada Ruiz; Predicting optimal treatment regimen for patients with neovascular age-related macular degeneration (nAMD) using machine learning. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2986 – F0256.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : This pilot study aimed to develop machine learning models to predict the optimal dosing regimen for patients with nAMD using baseline (BL) characteristics for patients treated with faricimab, the first bispecific antibody for intraocular use, in the phase 2 AVENUE (NCT02484690) and STAIRWAY (NCT03038880) trials.

Methods : Predicting optimal regimen, defined as the least frequent dosing regimen that achieves the maximum best-corrected visual acuity (BCVA) potential, is converted to a regression problem on the BCVA ETDRS letter score using treatment regimen as input. Because treatment arm is randomly assigned in clinical trials, we can alternate the treatment regimen feature to predict BCVA letter score for different regimens. To predict an optimal regimen for a new patient, BL characteristics are entered into the trained model together with each possible treatment regimen (Figure 1).
For the regression problem, BCVA at month 9 was set as the target. Linear model, random forest, extreme gradient boosting (XGBoost), and support vector machine are trained and evaluated. Age, sex, BCVA, and central subfield thickness at BL are selected as benchmark features. In addition, image-derived features, such as fluids and layer thickness, obtained by an automated segmentation algorithm on SD-OCT scans, are considered for full models. To evaluate the model’s discriminative power between treatment regimens, we define a metric called mean difference, which is the mean difference between the maximum and minimum prediction over all possible treatment regimen assignments.
The methods are applied to a data set consisting of all patients in AVENUE and STAIRWAY, including patients treated with ranibizumab (N = 324).

Results : For the regression problem, XGBoost achieved the highest performance (R2 = 0.38) with benchmark features and image features (Table 1). The mean difference value varies between models, with no obvious relationship between model performance and discriminative capability of treatment regimens.

Conclusions : The results of this study highlight the potential of a method to predict an optimal treatment regimen for patients with nAMD using a regression model. To fully understand advantages and limitations of this method, validation at a larger scale is warranted.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Figure 1. An Imaginary Scenario of Predicting an Optimal Regimen for a Patient

Figure 1. An Imaginary Scenario of Predicting an Optimal Regimen for a Patient

 

Table 1. Nested Cross-Validation

Table 1. Nested Cross-Validation

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×