icdcodex package

Subpackages

icdcodex.data package
- Module contents

Submodules

icdcodex.datacleaning module

icdcodex.hierarchy module

deserialize icd hierarchies computed in datacleaning.py

icdcodex.hierarchy.icd10cm(version: str | None = None) → Tuple[Graph, Sequence[str]][source]

deserialize icd-10-cm hierarchy

Parameters:: version (str, optional) – icd-10-cm version, including 2019 to 2024. If None, use the system year. Defaults to None.
Returns:: ICD-10-CM hierarchy and codes
Return type:: Tuple[nx.Graph, Sequence[str]]

icdcodex.hierarchy.icd9() → Tuple[Graph, Sequence[str]][source]

deserialize icd9 hierarchy

Returns:: ICD9 hierarchy and codes
Return type:: Tuple[nx.Graph, Sequence[str]]

icdcodex.icd2vec module

Build a vector embedding from a networkX representation of the ICD hierarchy

class icdcodex.icd2vec.Icd2Vec(num_embedding_dimensions: int = 128, num_walks: int = 10, walk_length: int = 10, window: int = 4, workers=1, **kwargs)[source]

Bases: object

fit(icd_hierarchy: Graph, icd_codes: Sequence[str], **kwargs)[source]

construct vector embedding of all ICD codes

Parameters:

icd_hierarchy (nx.Graph) – Graph of ICD hierarchy
kwargs – arguments passed to the Node2Vec.fit

to_code(vecs: Sequence[Sequence] | ndarray) → Sequence[str][source]

decode continuous representation of ICD code(s) into the code itself

Parameters:: vecs (Union[Sequence[Sequence], np.ndarray]) – continuous representation of ICD code(s)
Returns:: ICD code(s)
Return type:: Sequence[str]

to_vec(icd_codes: Sequence[str]) → ndarray[source]

encode ICD code(s) into a matrix of continuously-valued representations of shape m x n where m = self.num_embedding_dimensions and n = len(icd_codes)

Parameters:: icd_codes (Sequence[str]) – list of icd code(s)
Raises:: ValueError – If model is not fit beforehand
Returns:: continuously-valued representations if ICD codes
Return type:: np.ndarray