"""CirrDecompUnit-Kor Streamlit dashboard. Run: streamlit run app.py Standalone offline. Uses synthetic CSVs from data/. For research / QI use only. NOT for clinical decision making. """ from __future__ import annotations import os import sys from collections import defaultdict from typing import Dict, List HERE = os.path.dirname(os.path.abspath(__file__)) sys.path.insert(0, HERE) DATA_DIR = os.path.join(HERE, "data") try: import streamlit as st import pandas as pd import plotly.express as px except Exception as e: # pragma: no cover print("Streamlit/pandas/plotly are required. Install requirements.txt.") print(e) sys.exit(1) from modules import kpi_report, lt_candidacy # noqa: E402 # --------------------------------------------------------------------------- @st.cache_data def load_data() -> Dict[str, pd.DataFrame]: def _read(name): path = os.path.join(DATA_DIR, name) return pd.read_csv(path) if os.path.exists(path) else pd.DataFrame() return { "patients": _read("patients.csv"), "episodes": _read("episodes.csv"), "labs": _read("lab_trajectory.csv"), "adherence": _read("protocol_adherence.csv"), "lt": _read("lt_waitlist.csv"), "outcomes": _read("outcomes.csv"), } def main(): st.set_page_config(page_title="CirrDecompUnit-Kor", layout="wide") st.title("시르데컴프유닛코어 (CirrDecompUnit-Kor)") st.caption("MASLD/MASH cirrhosis decompensation inpatient QI dashboard " "— 참고용·연구용 (research / QI only)") data = load_data() if data["episodes"].empty: st.warning("No data found. Run: `python3 main.py --gen`") return epi = data["episodes"].copy() with st.sidebar: st.header("Filters") etio_opts = sorted(epi["etiology"].dropna().unique()) etio_sel = st.multiselect("Etiology", etio_opts, default=etio_opts) type_opts = sorted(epi["decomp_type"].dropna().unique()) type_sel = st.multiselect("Decompensation type", type_opts, default=type_opts) aclf_opts = ["no ACLF", "ACLF-1", "ACLF-2", "ACLF-3"] aclf_sel = st.multiselect("ACLF grade", aclf_opts, default=aclf_opts) ward_opts = sorted(epi["ward"].dropna().unique()) ward_sel = st.multiselect("Ward", ward_opts, default=ward_opts) flt = epi[ epi["etiology"].isin(etio_sel) & epi["decomp_type"].isin(type_sel) & epi["aclf_grade"].isin(aclf_sel) & epi["ward"].isin(ward_sel) ] if flt.empty: st.warning("No episodes after filtering.") return # ------------------------------------------------------------------ KPI c1, c2, c3, c4, c5 = st.columns(5) c1.metric("환자", flt["patient_id"].nunique()) c2.metric("Episodes", len(flt)) c3.metric("MELD 3.0 평균", f"{flt['meld3'].mean():.1f}") c4.metric("30일 사망률", f"{flt['mortality_30d'].mean():.1%}") c5.metric("90일 재입원율", f"{flt['readmission_90d'].mean():.1%}") tab1, tab2, tab3, tab4, tab5 = st.tabs([ "1. Cohort / etiology", "2. ACLF / MELD trajectory", "3. Protocol adherence", "4. LT candidacy / KONOS", "5. Post-discharge & report", ]) # 1. Cohort / etiology -------------------------------------------------- with tab1: st.subheader("Etiology mix") etio_ct = flt["etiology"].value_counts().reset_index() etio_ct.columns = ["etiology", "n"] st.plotly_chart(px.bar(etio_ct, x="etiology", y="n", color="etiology"), use_container_width=True) st.subheader("Decompensation type") d_ct = flt["decomp_type"].value_counts().reset_index() d_ct.columns = ["decomp_type", "n"] st.plotly_chart(px.bar(d_ct, x="decomp_type", y="n", color="decomp_type"), use_container_width=True) st.subheader("Per-ward episode load") ward_ct = flt.groupby(["ward", "decomp_type"]).size().reset_index(name="n") st.plotly_chart(px.bar(ward_ct, x="ward", y="n", color="decomp_type"), use_container_width=True) # 2. ACLF / MELD -------------------------------------------------------- with tab2: st.subheader("EASL-CLIF ACLF grade distribution") order = ["no ACLF", "ACLF-1", "ACLF-2", "ACLF-3"] g_ct = flt["aclf_grade"].value_counts().reindex(order, fill_value=0).reset_index() g_ct.columns = ["aclf_grade", "n"] st.plotly_chart(px.bar(g_ct, x="aclf_grade", y="n", color="aclf_grade", category_orders={"aclf_grade": order}), use_container_width=True) st.subheader("30-day mortality by ACLF grade") mort = flt.groupby("aclf_grade")["mortality_30d"].mean().reindex(order) mort_df = mort.reset_index().rename(columns={"mortality_30d": "rate"}) st.plotly_chart(px.bar(mort_df, x="aclf_grade", y="rate", labels={"rate": "30d mortality"}), use_container_width=True) st.subheader("MELD / MELD-Na / MELD 3.0 distribution") long = flt[["meld", "meld_na", "meld3"]].melt(var_name="score", value_name="value") st.plotly_chart(px.violin(long, x="score", y="value", box=True), use_container_width=True) st.subheader("CLIF-SOFA daily trajectory (sample episode)") if not data["labs"].empty: sample_eid = flt.iloc[0]["episode_id"] traj = data["labs"][data["labs"]["episode_id"] == sample_eid] if not traj.empty: st.plotly_chart(px.line(traj, x="day", y="clif_sofa", markers=True, title=f"Episode {sample_eid}"), use_container_width=True) # 3. Protocol adherence ------------------------------------------------- with tab3: st.subheader("Protocol adherence by decomp type") adh = flt.groupby("decomp_type")["protocol_adherence"].mean().reset_index() st.plotly_chart(px.bar(adh, x="decomp_type", y="protocol_adherence", color="decomp_type"), use_container_width=True) st.subheader("Adherence vs 30-day mortality") agg = flt.groupby("decomp_type").agg( adherence=("protocol_adherence", "mean"), mortality=("mortality_30d", "mean"), n=("episode_id", "count"), ).reset_index() st.plotly_chart(px.scatter(agg, x="adherence", y="mortality", size="n", color="decomp_type", text="decomp_type"), use_container_width=True) st.subheader("Ward ranking (mean adherence)") rank = (flt.groupby("ward")["protocol_adherence"].mean() .sort_values(ascending=False).reset_index()) st.dataframe(rank.style.format({"protocol_adherence": "{:.2%}"}), use_container_width=True) # 4. LT candidacy ------------------------------------------------------- with tab4: st.subheader("KONOS priority band") if not data["lt"].empty: lt = data["lt"].merge(flt[["episode_id"]], on="episode_id") band_ct = lt["konos_band"].value_counts().reset_index() band_ct.columns = ["konos_band", "n"] st.plotly_chart(px.pie(band_ct, names="konos_band", values="n"), use_container_width=True) st.dataframe(lt.head(50), use_container_width=True) st.markdown("MELD 3.0 ≥ 15 또는 ACLF-1/2/3 → LT 평가 대상. " "OPTN MELD 3.0 (2023-07) 및 KONOS 가이드 호환 (proxy).") # 5. Post-discharge & report ------------------------------------------- with tab5: st.subheader("Post-discharge follow-up rate") if not data["outcomes"].empty: o = data["outcomes"].merge(flt[["episode_id"]], on="episode_id") visits = { "POD7": o["pod7_visit"].mean(), "POD30": o["pod30_visit"].mean(), "POD90": o["pod90_visit"].mean(), } v_df = pd.DataFrame({"visit": list(visits), "rate": list(visits.values())}) st.plotly_chart(px.bar(v_df, x="visit", y="rate"), use_container_width=True) st.subheader("Discharge medications") med_cols = ["nsbb_on_discharge", "diuretic_on_discharge", "rifaximin_on_discharge", "sbp_prophylaxis_on_discharge", "resmetirom_on_discharge"] med_df = pd.DataFrame({"med": med_cols, "rate": [o[c].mean() for c in med_cols]}) st.plotly_chart(px.bar(med_df, x="med", y="rate"), use_container_width=True) st.subheader("Generate KASL/KLTS-style QI report") records = flt.to_dict("records") kpis = kpi_report.episode_kpis(records) lang = st.radio("언어", ["kor", "eng"], horizontal=True) st.code(kpi_report.render_text_report(kpis, lang=lang), language="markdown") st.divider() st.caption("Synthetic data only. de-identification: PHI columns (name/RRN/MRN) " "dropped at ingest. 참고용·연구용 only — NOT for clinical decision making.") if __name__ == "__main__": main()