Internal transfers are counted as readmissions

Aberta
  • Algo errado?
#1,269 3 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 2 dias

@thebean27 já está trabalhando nisso.

Desde 6/10/2026.

  • #1275 de @thebean27 — aberto

Avaliação

Dificuldade
4/5
Tempo estimado
3-5 dias
Facilidade para iniciantes
35/100
Tipo de issue
Bug
Clareza
Razoavelmente clara
Status de atividade
Estagnada
Stack de tecnologia
python
Domínio
data

Direção de pesquisa

The issue provides a reproducer using admissions.csv.gz and points to the prior exclusion clause removed in PR #815; start by tracing the readmission-label logic and reviewing PR #815 and the open PR #1275. Compare how admissions less than an hour apart are labeled with the intended treatment of internal transfers. Done means the behavior is addressed or its intended configuration is clarified, with tests covering the relevant case.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

Hi, first of all thank you for your repo; it has been useful for my research and very easy to use.

I have found an issue wherein internal transfers are counted as readmissions rather than treated as part of the same visit; which I think is surprising and probably not what most people want to model.

Reproduce (needs only admissions.csv.gz)

import polars as pl
pl.Config.set_fmt_str_lengths(60)   # print full admission_location strings

adm = (
    pl.read_csv("data/mimiciv/3.1/hosp/admissions.csv.gz", try_parse_dates=True)
       .sort("subject_id", "admittime")
       .with_columns(
              next_admit=pl.col("admittime").shift(-1).over("subject_id"),
              next_loc=pl.col("admission_location").shift(-1).over("subject_id")
       )
       .filter(pl.col("next_admit").is_not_null())
)
gap = pl.col("next_admit") - pl.col("dischtime")
pos = adm.filter(gap < pl.duration(days=30))          # the task's label rule
split = pos.filter(gap < pl.duration(hours=1))

print("total 30-day readmissions:", len(pos))     # 110k
print("readmissions within 1hr:  ", len(split))   # 10k (~9%)

# second admisison_location
print(split["next_loc"].value_counts(sort=True).head(5))
# ┌────────────────────────────────────────┬───────┐
# │ next_loc                               ┆ count │
# │ ---                                    ┆ ---   │
# │ str                                    ┆ u32   │
# ╞════════════════════════════════════════╪═══════╡
# │ INTERNAL TRANSFER TO OR FROM PSYCH     ┆ 5609  │
# │ PHYSICIAN REFERRAL                     ┆ 1844  │
# │ EMERGENCY ROOM                         ┆ 1451  │
# │ TRANSFER FROM SKILLED NURSING FACILITY ┆ 535   │
# │ WALK-IN/SELF REFERRAL                  ┆ 103   │
# └────────────────────────────────────────┴───────┘

Possibly related to https://patch-diff.githubusercontent.com/raw/sunlabuiuc/PyHealth/pull/815.diff
which removed a clause excluding admissions within 3hrs of a prev. admission.

-                if time_diff_hour <= 3:
-                    continue

A possible fix could be to merge admissions occuring within some time period and adding that as a configuration option and to emit some warning suggesting user's configure that value. I would imagine waiting for a major version change to change the default behaviour would be a good idea.

Environment: PyHealth 2.0.2, MIMIC-IV 3.1.

Linguagem predominante
Python
Estrelas
1.7k
Forks
803
Merge médio
4d 18h
PRs com merge (30d)
58

Preparar o ambiente

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de sunlabuiuc/PyHealth

Todas as issues de sunlabuiuc/PyHealth

Issues semelhantes

Mais issues de Python

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.