Patient-Trial Matching 

Table of Contents

Patient-Trial Matching
- Patient-Trial Matching: Index
- Patient-Trial Matching: Example

The patient-trial matching is a process that identifies patients who are eligible to participate in a clinical trial. The matching process is based on the patient’s electronic healthcare records (EHRs) and the trial’s inclusion/exclusion criteria. To formulate it as a machine learning problem, we first need to define the input and output of the matching process.

A patient’s EHR can be represented by a sequence of visits $V = \{v_1, v_2, ..., v_n\}$ , where each visit $v_i$ is a sequence of events $v_i = \{o_{i1}, o_{i2}, ..., o_{im}\}$ . We can encode $V$ to a compact embedding $E_p \in R^d$ as the patient’s representation. The trial’s inclusion/exclusion criteria can be represented by a sequence of inclusion/exclusion criteria $C = \{c_1, c_2, ..., c_m\}$ . We can encode $C$ to a compact embedding $E_c \in R^{m\times d}$ the trial’s representation (each criterion is a vector). Such that, the neural matching process is just compute the similarity between $E_p$ and $E_c^m$ to get the criterion-level affinity.