Table 1.

Examples of discrete token creation alongside percentile-based quantization and tokenization of continuous variables.

Label	Value	Token
Eye opening	4 Spontaneously	Eye Opening_4 Spontaneously
Heart rate	84	Heart Rate_8
Code status	Full code	Code Status_Full Code
Systolic blood pressure	NaN	Systolic blood pressure_NaN

Discrete variables remain discrete, while values that can be converted to floating point numbers are considered continuous and separated into percentile-based bins. Our pipeline is simply: convert non-NaN numeric values to percentiles, everything else is considered discrete.