"""Streamlit dashboard — BariERASRecov-Kor. 비만수술 perioperative ERAS recovery cohort dashboard. Run: pip install -r requirements.txt streamlit run app.py For research / synthetic data only. NOT for clinical decision making. """ from __future__ import annotations import os import sys DISCLAIMER = ("본 대시보드는 합성 데이터 기반 참고용·연구용입니다. " "임상 의사결정에 직접 사용하지 마십시오. " "(For research / synthetic data only — NOT for clinical decision.)") def _missing_deps_panel(err: Exception): print(f"[BariERASRecov-Kor] streamlit 의존성 누락: {err}", file=sys.stderr) print("가상환경에서 `pip install -r requirements.txt` 후 다시 실행하세요.", file=sys.stderr) sys.exit(0) def main() -> None: try: import streamlit as st import pandas as pd import plotly.express as px import plotly.graph_objects as go except ImportError as e: _missing_deps_panel(e) return here = os.path.dirname(os.path.abspath(__file__)) sys.path.insert(0, here) from modules import ingest, eras, procedure, hypo, outpatient st.set_page_config( page_title="BariERASRecov-Kor", page_icon="🏥", layout="wide", ) st.title("BariERASRecov-Kor — 비만수술 perioperative ERAS recovery dashboard") st.caption("RYGB · SG · OAGB · SADI · DJB | ERAS Society · MBSAQIP · " "ASMBS · KASMBS · KDA") st.warning(DISCLAIMER) # -- Sidebar: data load ----------------------------------------------- st.sidebar.header("데이터") data_dir = st.sidebar.text_input("CSV 디렉토리", value=os.path.join(here, "data")) if st.sidebar.button("Synthetic data 재생성 (n=400, seed=42)"): with st.spinner("생성 중..."): ingest.generate_synthetic(n_patients=400, out_dir=data_dir, seed=42) st.sidebar.success("data/ 재생성 완료") if not os.path.exists(os.path.join(data_dir, "patients.csv")): st.error("patients.csv 없음. 사이드바에서 synthetic data 생성 또는 " "`python3 main.py --gen-data` 실행하세요.") return patients, intraop, pod03, pod430, pod90, hypo_ev, irep = ingest.load_all(data_dir) # Sidebar filters st.sidebar.header("필터") procs = sorted({p.procedure for p in patients}) sel_procs = st.sidebar.multiselect("술식", procs, default=procs) wards = sorted({p.ward for p in patients}) sel_wards = st.sidebar.multiselect("Ward", wards, default=wards) patients_f = [p for p in patients if p.procedure in sel_procs and p.ward in sel_wards] keep = {p.patient_id for p in patients_f} intraop_f = [r for r in intraop if r.patient_id in keep] pod03_f = [r for r in pod03 if r.patient_id in keep] pod430_f = [r for r in pod430 if r.patient_id in keep] pod90_f = [r for r in pod90 if r.patient_id in keep] hypo_f = [r for r in hypo_ev if r.patient_id in keep] # -- KPI cards -------------------------------------------------------- c1, c2, c3, c4, c5 = st.columns(5) c1.metric("환자 N", len(patients_f)) c2.metric("Mean BMI", round( sum(p.bmi_pre for p in patients_f) / max(1, len(patients_f)), 1)) n_leak = sum(1 for r in pod03_f if r.leak or r.staple_leak) c3.metric("30d Leak %", round(100 * n_leak / max(1, len(patients_f)), 2)) n_readmit = sum(1 for r in pod430_f if r.readmit_30d) c4.metric("30d Readmission %", round(100 * n_readmit / max(1, len(patients_f)), 2)) n_mort = sum(1 for p in patients_f if p.died_30d) c5.metric("30d Mortality %", round(100 * n_mort / max(1, len(patients_f)), 2)) # -- Tabs ------------------------------------------------------------- tabs = st.tabs([ "1) ERAS bundle radar", "2) MBSAQIP measures", "3) Procedure stratification", "4) Hypo + dumping", "5) POD0-90 outpatient", "6) KM survival", "7) Raw data", ]) # 1) ERAS radar with tabs[0]: st.subheader("ERAS Society 비만수술 protocol — ward별 radar") bundles = eras.compute_patient_bundles( patients_f, intraop_f, pod03_f, pod430_f) radar = eras.ward_radar(bundles) rdf = pd.DataFrame([{ "ward": w.ward, "N": w.n_patients, "preop": w.preop_pct, "intraop": w.intraop_pct, "POD0-3": w.pod03_pct, "POD4-30": w.pod430_pct, "overall": w.overall_pct, } for w in radar]) st.dataframe(rdf, use_container_width=True) fig = go.Figure() categories = ["preop", "intraop", "POD0-3", "POD4-30"] for w in radar: fig.add_trace(go.Scatterpolar( r=[w.preop_pct, w.intraop_pct, w.pod03_pct, w.pod430_pct], theta=categories, fill="toself", name=w.ward, )) fig.update_layout( polar=dict(radialaxis=dict(visible=True, range=[0, 100])), showlegend=True, height=500, ) st.plotly_chart(fig, use_container_width=True) # 2) MBSAQIP with tabs[1]: st.subheader("MBSAQIP / ASMBS analog quality measures") m = eras.mbsaqip_measures(patients_f, pod03_f, pod430_f) mdf = pd.DataFrame([{ "measure": x.measure, "rate %": x.rate_pct, "events": x.n_events, "N": x.n_denominator, "target %": x.target_pct, "flag": ("PASS" if x.rate_pct <= x.target_pct else "WATCH"), } for x in m]) st.dataframe(mdf, use_container_width=True) fig = px.bar(mdf, x="measure", y="rate %", color="flag", text="rate %", color_discrete_map={"PASS": "#4CAF50", "WATCH": "#E53935"}) fig.update_layout(xaxis_tickangle=-30, height=420) st.plotly_chart(fig, use_container_width=True) # 3) Procedure stratification with tabs[2]: st.subheader("5+ 술식별 stratification") rows = procedure.stratify_by_procedure(patients_f, pod03_f, pod430_f) df = pd.DataFrame([r.__dict__ for r in rows]) st.dataframe(df, use_container_width=True) if not df.empty: fig = px.bar(df, x="procedure", y=["leak_rate_pct", "marginal_ulcer_pct", "dumping_any_pct", "readmit_30d_pct"], barmode="group", height=420, labels={"value": "% of N", "variable": "endpoint"}) st.plotly_chart(fig, use_container_width=True) oe_cols = ["oe_leak", "oe_readmit", "oe_mortality"] oe_df = df[["procedure"] + oe_cols].copy() st.markdown("**Risk-adjusted O/E ratio (vs. literature anchor)**") st.dataframe(oe_df, use_container_width=True) # 4) Hypoglycemia / dumping with tabs[3]: st.subheader("post-bariatric hypoglycemia + dumping syndrome 시간 분포") bucket = hypo.time_distribution(hypo_f) bdf = pd.DataFrame([r.__dict__ for r in bucket]) st.dataframe(bdf, use_container_width=True) if not bdf.empty: fig = px.bar(bdf, x="time_bucket", y="n_events", color="hypo_type", barmode="stack", height=380) st.plotly_chart(fig, use_container_width=True) pdf = pd.DataFrame([r.__dict__ for r in hypo.by_procedure( patients_f, hypo_f)]) st.markdown("**By procedure**") st.dataframe(pdf, use_container_width=True) # 5) POD0-90 outpatient with tabs[4]: st.subheader("POD0-90 outpatient follow-up & transition") adh = outpatient.adherence_by("procedure", patients_f, pod90_f) adh_df = pd.DataFrame([r.__dict__ for r in adh]) st.dataframe(adh_df, use_container_width=True) if not adh_df.empty: long_df = adh_df.melt( id_vars=["key"], value_vars=["pct_pod7", "pct_pod30", "pct_pod60", "pct_pod90"], var_name="visit", value_name="adherence_pct") fig = px.bar(long_df, x="key", y="adherence_pct", color="visit", barmode="group", height=420, labels={"key": "procedure"}) st.plotly_chart(fig, use_container_width=True) trans = outpatient.transition_summary(patients_f, pod430_f) st.markdown(f"**30-day readmission**: {trans.n_readmit_30d}/" f"{trans.n_total} ({trans.readmit_rate_pct}%) | " f"**mortality**: {trans.n_mort_30d} " f"({trans.mort_rate_pct}%) | " f"median readmit day: {trans.median_time_to_readmit_d}") st.write("readmit reason mix:", trans.readmit_reason_mix) # 6) KM survival with tabs[5]: st.subheader("Kaplan-Meier (30-day readmission)") km = outpatient.kaplan_meier_step( patients_f, pod430_f, endpoint="readmit", horizon_d=30) kdf = pd.DataFrame(km, columns=["day", "S(t)", "n_at_risk"]) st.dataframe(kdf, use_container_width=True) fig = px.line(kdf, x="day", y="S(t)", height=380, title="30-day survival free of readmission") fig.update_layout(yaxis=dict(range=[0.0, 1.0])) st.plotly_chart(fig, use_container_width=True) # 7) Raw with tabs[6]: st.subheader("Raw data (de-identified)") st.markdown(f"**de-id**: {irep.deid_method}") st.markdown("**patients** (head)") st.dataframe(pd.DataFrame([p.__dict__ for p in patients_f[:50]]), use_container_width=True) st.markdown("**hypo events** (head)") st.dataframe(pd.DataFrame([e.__dict__ for e in hypo_f[:50]]), use_container_width=True) if __name__ == "__main__": main()