EHR Intelligence

EHR Data Representation

Temporal Event Sequences

import pandas as pd
import numpy as np

class EHRProcessor:
    def __init__(self, code_system='ICD10'):
        self.code_system = code_system

    def temporal_features(self, encounters_df):
        features = {}
        for pid, group in encounters_df.groupby('patient_id'):
            features[pid] = {
                'visit_count': len(group),
                'unique_codes': group['code'].nunique(),
                'time_span': (group['date'].max() - group['date'].min()).days,
                'avg_interval': group['date'].diff().mean().days
            }
        return pd.DataFrame(features).T

Predictive Modeling

Attention-Based Visit Prediction

class EHRTransformer(nn.Module):
    def __init__(self, n_codes, d_model=256, n_heads=8, n_layers=4):
        super().__init__()
        self.code_embedding = nn.Embedding(n_codes + 1, d_model)
        encoder_layer = nn.TransformerEncoderLayer(
            d_model=d_model, nhead=n_heads, dim_feedforward=512)
        self.transformer = nn.TransformerEncoder(encoder_layer, n_layers)
        self.classifier = nn.Linear(d_model, 1)

    def forward(self, visit_codes, visit_mask):
        x = self.code_embedding(visit_codes).permute(1, 0, 2)
        h = self.transformer(x, src_key_padding_mask=visit_mask)
        return self.classifier(h.mean(dim=0))

Clinical Event Prediction

Comorbidity Indices