Treatment effect prediction (TEP) plays a vital role in disease management by ensuring that the expected clinical outcomes are obtained after performing specialized and sophisticated treatments on patients given their personalized clinical status. To address this problem, we propose an adversarial deep treatment effect prediction model by utilizing the potential of a large volume of electronic health records (EHR) data. Our model employs two auto-encoders for learning the representative and discriminative features of both patient characteristics and treatments from EHR data. The discriminative power of the learned features is further enhanced by decoding the correlational information between the patient characteristics and subsequent treatments by mean s of a generative adversarial learning strategy. Thereafter, a logistic regression layer is appended on the top of the resulting feature representation layer for TEP. The proposed model was evaluated on a real clinical dataset and the experimental results demonstrate that our proposed model achieves competitive performance compared to state-of-the-art models in tackling the TEP problem.
Treatment effect prediction (TEP), as ensuring to obtain the expected clinical outcomes
after performing specialized and sophisticated treatments on patients given their
personalized clinical status, is vital for disease management. Traditional approaches to
addressing this problem have mostly relied on randomized controlled trial (RCT) studies
[
Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
Electronic health records (EHRs), with their increasingly widespread adoption in
clinical practice, provide a comprehensive source for treatment effect analysis to
augment traditional RCT studies [
In this study, we propose a novel adversarial deep treatment effect prediction
(ADTEP) model to anticipate treatment effects by utilizing a large volume of EHR data.
In detail, two Auto-encoders (AE) are employed to encode the physical condition and
treatment information of patient samples into latent robust representations. To align the
generated treatments with the actual performed treatments, we adopt an adversarial
learning scheme [
We consider a typical clinical study of TEP, in which the EHR data record patient features, treatment interventions, and achieved treatment outcomes. For each patient sample , we observe a set of patient features , a set of treatment interventions conditioned on , and the achieved treatment outcome . The EHR dataset can be described as , , | 1, ⋯ , . We propose the ADTEP model to adare concatenated to form the input of C for TEP. dress the aforementioned problem. The proposed ADTEP contains seven components: a patient feature encoder E , a treatment intervention encoder E , a patient feature decoder G , a treatment intervention decoder G , a treatment intervention generator G , a treatment intervention discriminator D , and a logistic regression layer for TEP C . In detail, given a patient sample , , to extract the latent features and , two encoder layers E and E are first employed
from and , respectively. The reconstructed features and can then be estimated from the latent features adopted to capture robust and discriminative patient feature/treatment representations , using the in the latent feature vector / . Consequently, the latent feature vectors and
We measure the reconstruction performance for patient feature conducted by the encoder E
and decoder G . For efficient learning of the encoder-decoder, standard practice is to use the Euclidean distance between the input and the generated output to minimize the patient feature reconstruction loss, that is, . treatment reconstruction loss ℒ can be measured as follows: The reconstruction performance for treatment vector is measured by means of the encoder E and decoder G . Similarly to the patient feature reconstruction loss ℒ , the .
To encourage the reconstruction of treatments from discriminative patient features that are similar to real ones, so that the prediction performance can be enriched, we design a treatment discriminator D to differentiate the reconstructed treatment vector from the true observed treatment . In particular, we employ a binary classifier to categorize the given input as “real” if the input is the actual treatment vector performed on patients, and “fake” otherwise. The adversarial loss ℒ is defined as: Given a testing patient sample with patient feature vector , treatment vector conditioned on , and an unknown treatment outcome label y, we can learn the representative and informative features and
with respect to the patient characteristics, and subsequently the treatments performed on the patient, respectively, and then concatenate these as ,
to be fed into the treatment effect predictor C . Let treatment outcome, the loss can be measured using cross-entropy as follows: is the predicted struction ℒ , and loss of treatment outcome prediction ℒ tive function of the ADTEP is expressed as: As demonstrated in the section above, our training is defined by four loss functions: 1) loss of GAN ℒ , loss of patient feature reconstruction ℒ , loss of treatment recon. In summary, the objec, ing components. 3
where and are trade-off parameters for balancing the importance of the
correspondWe conducted a clinical case study in cooperation with the Cardiology Department of
the Chinese PLA General Hospital. The primary investigated major adverse event
prediction (MACE) after acute coronary syndrome (ACS). ACS refers to a group of
conditions resulting from decreased blood flow in the coronary arteries, whereby that part
of the heart muscle is unable to function properly or dies [
Precision
To demonstrate the effectiveness of our proposed model, we compare the proposed ADTEP with the proposed model without adversarial learning, namely the DTEP model. For the DTEP, we use AEs to generate the latent representations of both the patient characteristics and the subsequent treatments, concatenate the derived latent features, and then feed the obtained feature vector into a logistic regression layer, yielding a TEP model. Moreover, we compare the proposed model to state-of-the-art models using the experimental datasets, including logistic regression (LR) and the support vector machine (SVM).
The performance was evaluated by the Area Under the receiver operating characteristic (ROC) curve (AUC), accuracy, precision, recall and F1 score. We repeated the experiments five times to validate the performance of each model on the experimental dataset. As a result, we obtained a group of experimental results for each model, on which the mean value and confidence intervals were calculated.
In this work, we have addressed quite a challenging problem in medical informatics, namely utilizing a large volume of observational data for TEP. Our proposed model was evaluated on a real clinical dataset, and the experimental results demonstrate significant improvements in TEP compared to state-of-the-art methods.
This work was partially supported by the National Key Research and Development Program of China under Grant No. 2016YFC1300303 and the National Nature Science Foundation of China under Grant No. 61672450.