JMIR Publications

ABSTRACT

Background:

Irregularly sampled data, as a common data structure in the medical field, poses problems such as unequal sampling time intervals and frequencies, making it difficult to align the data without losing information for input into models. Meanwhile, models also struggle to fit complex tasks due to the loss of information in the data.

Objective:

To address the impact of issues such as missing values and irregularly sampling time intervals on prediction results, this paper proposes a method that converts time series data into a graph network structure and develops a prediction model for graph data to better support complex tasks.

Methods:

In this paper, feature channels and measurement times in time series data are constructed as nodes, and feature measurements at time points are constructed as edge weights. The point feature space and edge weight space are extracted by graph convolutional neural network and gated convolutional attention mechanism network respectively, and the feature fusion is carried out under the action of the learnable head.

Results:

The proposed method is tested on four datasets and compared with advanced time series prediction models, analyzing data from five key clinical treatment stages. In both classification and regression tasks, our method outperforms others in most experiments(ACC=0.6775-0.9200, MAE=0.0355-0.0646). Overall, the results show that our method can stably and accurately predict multiple downstream tasks (Mean ACC=0.737-0.906).

Conclusions:

The proposed model can effectively analyze medical data with missing values and uneven sampling. It achieves high prediction accuracy in both regression prediction tasks and classification tasks. The graph network-based model structure endows the model with better interpretability, enabling effective exploration of the factors that trigger changes.