Predictive Clinical Decision Support System with RNN Encoding and Tensor Decoding

Yinchong Yang, Peter A. Fasching, Markus Wallwiener, Tanja N. Fehm, Sara Y. Brucker, Volker TresparXiv 2016 · arXiv

TL;DR

This research presents a novel machine learning architecture for Clinical Decision Support (CDS) systems that predicts physician treatment decisions by modeling them as joint probabilities. By integrating a Recurrent Neural Network (RNN) encoder for longitudinal patient history with a tensor factorization decoder, the model captures complex dependencies between simultaneous clinical choices, significantly outperforming traditional independent classification models.

The Challenge: Modeling Correlated Clinical Decisions

In modern healthcare, Electronic Health Records (EHRs) provide a wealth of data, but translating this into actionable advice for physicians remains difficult. A primary limitation of current Clinical Decision Support (CDS) systems is their tendency to treat medical decisions as isolated events.

In practice, physicians often make "blocks" of mutually dependent decisions. For instance, when prescribing radiotherapy, a doctor simultaneously decides on the intention (e.g., curative vs. palliative) and the type (e.g., percutaneous vs. brachytherapy). Standard machine learning approaches typically predict these targets independently, calculating marginal probabilities and ignoring the inherent correlations between them. This paper addresses this gap by modeling the joint probability of these correlated clinical decisions.

Proposed Methodology: The RNN-Tensor Framework

The authors propose an end-to-end Encoder-Decoder architecture that combines sequential deep learning with multi-relational tensor algebra.

1. The Encoder (Patient Representation)

To handle the complexity of patient data, the model employs a dual-input strategy:

  • Sequential Data: An LSTM (Long Short-Term Memory) network processes variable-length clinical histories, including past therapies and clinical events.
  • Static Data: Baseline information, such as genetics and primary tumor details, is processed via a log-linear mapping. The outputs from both are concatenated into a single, fixed-size latent vector that represents the patient's current clinical state.

2. The Decoder (Tensor Factorization)

The core innovation lies in the decoder. Instead of standard classification layers, the model uses a Tucker-3 tensor factorization approach. By constructing a target tensor that represents the joint occurrence of decision pairs (e.g., Intention $\times$ Type) across patients and time, the model can learn the underlying structure of physician behavior.

The proposed architecture combining log-linear mapping for static data, an RNN encoder for sequential patient history, and a tensor factorization decoder for joint decision modeling.

Evaluation on Real-World Clinical Data

The model was rigorously tested using data from the PRAEGNANT study network, focusing on 1,245 patients with metastatic breast cancer. The system was tasked with predicting radiotherapy prescriptions, which involve two highly correlated dimensions: Therapy Intention and Therapy Type. Statistical analysis (Pearson’s $\chi^2$ and G-Tests) confirmed that these decisions are indeed strongly interdependent, justifying the joint modeling approach.

Key Results and Performance

The "Tensor Model" was compared against several baselines, including a "Standard Model" that predicts decisions independently.

  • Classification Accuracy: The Tensor Model achieved an AUROC of 0.874, a notable improvement over the standard model's 0.842.
  • Recommendation Quality: The most significant gains were observed in ranking metrics. In a CDS context, providing a ranked list of viable treatment combinations is often more valuable than a single prediction. The Tensor Model achieved an NDCG@5 of 0.269, representing a 56% improvement over the standard model (0.172).
Performance comparison across various metrics, showing the Tensor Model's superiority in both classification (AUROC) and ranking (NDCG@5).

Insights and Future Directions

The success of this architecture highlights two critical points for future CDS development:

  1. Joint Distributions Matter: Treating clinical decisions as a joint probability problem aligns more closely with real-world medical practice than independent classification.
  2. Tensors as Decoders: Tensor factorization, typically used in recommender systems, is an effective way to decode RNN latent states when the output space involves multi-dimensional, correlated categorical variables.

While currently focused on radiotherapy (a 3-D tensor problem), the authors plan to scale the architecture to handle more complex scenarios, such as systemic therapies, which involve even higher-dimensional decision tensors.

Summary updated

Abstract

With the introduction of the Electric Health Records, large amounts of digital data become available for analysis and decision support. When physicians are prescribing treatments to a patient, they need to consider a large range of data variety and volume, making decisions increasingly complex. Machine learning based Clinical Decision Support systems can be a solution to the data challenges. In this work we focus on a class of decision support in which the physicians' decision is directly predicted. Concretely, the model would assign higher probabilities to decisions that it presumes the physician are more likely to make. Thus the CDS system can provide physicians with rational recommendations. We also address the problem of correlation in target features: Often a physician is required to make multiple (sub-)decisions in a block, and that these decisions are mutually dependent. We propose a solution to the target correlation problem using a tensor factorization model. In order to handle the patients' historical information as sequential data, we apply the so-called Encoder-Decoder-Framework which is based on Recurrent Neural Networks (RNN) as encoders and a tensor factorization model as a decoder, a combination which is novel in machine learning. With experiments with real-world datasets we show that the proposed model does achieve better prediction performances.
BibTeX
@misc{yang2016predictive,
  author = {Yinchong Yang and Peter A. Fasching and Markus Wallwiener and Tanja N. Fehm and Sara Y. Brucker and Volker Tresp},
  title = {{Predictive Clinical Decision Support System with RNN Encoding and Tensor Decoding}},
  year = {2016},
  eprint = {1612.00611},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1612.00611},
}
Figure