"""Descriptive point-estimate audit of published SCREEN eTable 16 cells.

No confidence interval or causal identification is claimed: cluster-period data
are unavailable, and COVID status was measured anytime during hospitalization.
"""
import csv,json,pathlib
rows=list(csv.DictReader(open('data/screen_periods.csv')))
for r in rows:
    for k in ['period','n_alert','deaths_alert','n_control','deaths_control']:r[k]=int(r[k])
def totals(rr):return {k:sum(r[k] for r in rr) for k in ['n_alert','deaths_alert','n_control','deaths_control']}
def crude(rr):
    d=totals(rr)
    return (d['deaths_alert']/d['n_alert'])/(d['deaths_control']/d['n_control'])
def mh(rr):
    usable=[r for r in rr if r['n_alert']>0 and r['n_control']>0]
    a=sum(r['deaths_alert']*r['n_control']/(r['n_alert']+r['n_control']) for r in usable)
    b=sum(r['deaths_control']*r['n_alert']/(r['n_alert']+r['n_control']) for r in usable)
    return a/b
periods=[]
for p in range(1,11):periods.append(dict(period=p,**totals([r for r in rows if r['period']==p])))
all_counts=totals(rows)
assert all_counts==dict(n_alert=29442,deaths_alert=937,n_control=30613,deaths_control=961)
for p in periods:
    assert 0<=p['deaths_alert']<=p['n_alert'] and 0<=p['deaths_control']<=p['n_control']
overlap=[r for r in rows if 2<=r['period']<=9]
result=dict(source='Arabi et al. JAMA 2025 Supplement 2 eTable 16, PDF page 29',all_counts=all_counts,overlap_counts=totals(overlap),raw_rr=round(crude(rows),4),overlap_raw_rr=round(crude(overlap),4),period_mh_rr=round(mh(periods),4),period_covid_mh_rr=round(mh(rows),4),concurrent_periods=8,periods=10,omitted_nonoverlap_periods=[1,10],inference='Descriptive point contrasts only; no cluster-valid standard errors or causal identification')
pathlib.Path('results/R4.json').write_text(json.dumps(result,indent=2)+'\n')
print(json.dumps(result,indent=2))
