Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

OMOP: add observation_period and visit_detail to the default config

オープン
  • 何か問題がありますか？
#1,294 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 3 日以内に返信

まだ誰も着手していません。

評価

難易度
3/5
見積もり時間
1〜2日
初心者へのやさしさ
42/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
python
領域
data, machine-learning

調査の方向性

Start with the table list in pyhealth/datasets/configs/omop.yaml and the docstring in pyhealth/datasets/omop.py near lines 38-39, which promises observation and device_exposure. Add the missing table entries, then check that passing them no longer raises "Table ... not found in config". Before writing code, wait for maintainers to answer the two open questions (config version and task option versus filter), since they decide the shape of the change.

索引モデルが issue の本文から書いたものです。

説明

I am Koji Miyazawa, a data scientist based in Japan. I work with healthcare big data, including DPC inpatient data and Japanese claims data. I use Bayesian statistics and machine learning. I build data warehouses. I maintain jpomop (https://github.com/kojikojikojikoji/jpomop), which converts Japanese DPC data to the OMOP CDM.

What I found (main at b984552):

  1. The default OMOP config loads seven tables: person, visit_occurrence, death, condition_occurrence, procedure_occurrence, drug_exposure and measurement. It has no observation_period, visit_detail or observation.
    https://github.com/sunlabuiuc/PyHealth/blob/b9845528fe85ed3afb614a10b1f0b8ba32ef1668/pyhealth/datasets/configs/omop.yaml

  2. The OMOPDataset docstring says observation and device_exposure can be added. They are not in the config, so passing them raises "Table ... not found in config".
    https://github.com/sunlabuiuc/PyHealth/blob/b9845528fe85ed3afb614a10b1f0b8ba32ef1668/pyhealth/datasets/omop.py#L38-L39

  3. MortalityPredictionOMOP gives label 1 if the person has any death record, and 0 if not. It does not check how long the person was observed.
    https://github.com/sunlabuiuc/PyHealth/blob/b9845528fe85ed3afb614a10b1f0b8ba32ef1668/pyhealth/tasks/mortality_prediction.py#L1125-L1136

The CDM v5.4 docs say that outside observation_period, "absence of an Event ... cannot be construed as evidence of absence of an Event." A person who leaves the data early can get label 0 even when we do not know the outcome.
https://ohdsi.github.io/CommonDataModel/cdm54.html#observation_period

My proposal:

  • Add observation_period and visit_detail to omop.yaml. Maybe also add observation and device_exposure to match the docstring.
  • Optional: in tasks, use observation_period to drop samples without enough follow-up. Keep this off by default, so current results do not change.

I can test this with synthetic OMOP data converted from DPC by jpomop. jpomop builds observation_period. It also builds visit_detail from daily ward codes.

Questions:

  • Should the config stay at "5.3", or move to 5.4?
  • Is a task option fine, or do you prefer a separate filter?

Would a PR be welcome?

主要言語
Python
スター
1.7k
フォーク
803
平均マージ
5日 20時間
マージ済み PR（30日）
56

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

sunlabuiuc/PyHealth のほかの issue

sunlabuiuc/PyHealth の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。