CDC Adult Sepsis Event (ASE) Computation¶

Compute CDC Adult Sepsis Events from CLIF data using the CDC Sepsis Surveillance Toolkit (2018) criteria.

Sepsis = Component A (Presumed Serious Infection) + Component B (Acute Organ Dysfunction)

Component A: Presumed Serious Infection

Blood culture obtained (result not required)
≥4 Qualifying Antimicrobial Days (QAD) starting within ±2 calendar days of blood culture
- Must include new antimicrobial (not given in prior 2 days)
- At least one IV/IM antimicrobial required on day 1
- 1-day gaps allowed between antimicrobial days

Component B: Acute Organ Dysfunction

At least ONE within ±2 calendar days of blood culture:

Cardiovascular: New vasopressor initiation
Respiratory: New IMV initiation
Renal: Creatinine ≥2x baseline (ESRD excluded)
Hepatic: Bilirubin ≥2.0 mg/dL AND ≥2x baseline
Coagulation: Platelets <100 AND ≥50% decline
Metabolic: Lactate ≥2.0 mmol/L (optional)

Limitations

No eGFR criterion: CLIF does not include eGFR data, so the CDC's alternative renal dysfunction criterion (eGFR decrease) is not implemented—only creatinine doubling is used
Baseline selection: The CDC definition contains an inherent circular dependency: determining onset type requires the onset datetime, which requires organ dysfunction timing, which requires knowing the baseline, which depends on onset type. This implementation resolves the circularity by using blood culture timing (Day 1-2 vs Day 3+) to select which baseline to apply, then calculates the final onset type from the resulting ASE onset datetime. See Resolving the Circular Dependency for details.

For a comprehensive visual representation of the ASE algorithm:

🔍 View ASE Flow Diagrams

Installation and Import¶

from clifpy.utils.ase import compute_ase
# or
from clifpy.utils import compute_ase

# Compute ASE for specific hospitalizations
ase_results = compute_ase(
    hospitalization_ids=hosp_ids,
    data_directory='/path/to/clif/data',
    filetype='parquet',
    timezone='UTC'
)

# Or using a config file
ase_results = compute_ase(
    hospitalization_ids=hosp_ids,
    config_path='config/config.json'
)

Parameters¶

from clifpy.utils.ase import compute_ase

compute_ase(
    hospitalization_ids=None,      # List of hospitalization IDs (optional)
    config_path=None,              # Path to CLIF config file
    data_directory=None,           # Override data directory path
    filetype='parquet',           # Data format ('parquet' or 'csv')
    timezone='UTC',               # Timezone for datetime handling
    apply_rit=True,               # Apply 14-day Repeat Infection Timeframe
    rit_only_hospital_onset=True, # Apply RIT only to hospital-onset events
    include_lactate=False,        # Include lactate as organ dysfunction (default: False)
    verbose=True                  # Enable detailed logging (default: True)
)

Key Parameters¶

hospitalization_ids: Process specific hospitalizations or all if None
config_path: CLIF configuration file containing data paths and settings
apply_rit: Filters repeat infections within 14 days (per CDC guidelines)
rit_only_hospital_onset: When True, RIT only applies to hospital-onset events
include_lactate: Include lactate ≥2.0 as a qualifying organ dysfunction

Output¶

Returns a DataFrame with one row per blood culture episode containing:

Key Columns¶

Column	Description
`hospitalization_id`	Unique hospitalization identifier
`bc_id`	Blood culture index within hospitalization (1, 2, 3...)
`episode_id`	ASE episode number (after RIT filtering)
`type`	Onset type: 'community' (≤2 days) or 'hospital' (>2 days)
`presumed_infection`	Component A flag (1=meets criteria)
`sepsis`	Final ASE flag with lactate
`sepsis_wo_lactate`	ASE flag excluding lactate criterion
`no_sepsis_reason`	Why sepsis criteria not met

Timestamps¶

blood_culture_dttm: Blood culture collection/order time
ase_onset_w_lactate_dttm: ASE onset date (earliest qualifying event)
ase_first_criteria_w_lactate: First criterion triggering ASE
Various organ dysfunction timestamps

QAD Details¶

total_qad: Number of Qualifying Antimicrobial Days
qad_start_date: First QAD date
qad_end_date: Last consecutive QAD date
anchor_meds_in_window: New antimicrobials within ±2 days

Implementation Details¶

Baseline Calculation¶

Community baseline: Uses minimum (creatinine, bilirubin) or maximum (platelets) values across entire hospitalization
Hospital baseline: Uses values within ±2 days of blood culture
Selection based on blood culture timing (≤2 days = community, >2 days = hospital)

QAD Censoring Rules¶

Patients may meet QAD with <4 days if they:

Die (expired)
Transfer to acute care hospital
Enter hospice care

These events must occur within 3 days of QAD start, with antibiotics continuing until discharge.

Repeat Infection Timeframe (RIT)¶

For hospital-onset events, the 14-day RIT prevents duplicate counting:

Only the first event in each 14-day window is counted
Community-onset events are typically excluded from RIT
Each qualifying event gets a unique episode_id

ESRD Exclusion¶

Patients with End-Stage Renal Disease are excluded from AKI criterion. ESRD identified by ICD codes:

N18.6: End stage renal disease
Z49.31: Encounter for CRRT
Z49.01: Encounter for dialysis
I12.0, I13.11, I13.2: Hypertensive CKD with ESRD

Example: Detailed Usage¶

from clifpy.utils.ase import compute_ase, ase
import pandas as pd

# Load your cohort
cohort_df = pd.read_parquet('cohort.parquet')
hosp_ids = cohort_df['hospitalization_id'].unique().tolist()

# Compute ASE with all options
ase_results = compute_ase(
    hospitalization_ids=hosp_ids,
    config_path='config/config.json',
    apply_rit=True,                # Apply 14-day filtering
    rit_only_hospital_onset=True,  # RIT for hospital events only
    include_lactate=False,          # Exclude lactate criterion (default)
    verbose=True                    # Show detailed progress (default)
)

# Or use the alias function
ase_results = ase(
    hospitalization_ids=hosp_ids,
    data_directory='/path/to/clif/data',
    filetype='parquet',
    timezone='UTC'
)

# Analyze results
print(f"Total blood cultures evaluated: {len(ase_results)}")
print(f"Presumed infections: {ase_results['presumed_infection'].sum()}")
print(f"ASE sepsis events: {ase_results['sepsis'].sum()}")

# Get first ASE event per patient
first_ase = ase_results[
    (ase_results['sepsis'] == 1) &
    (ase_results['episode_id'] == 1)
]

# Examine onset types
print(ase_results[ase_results['sepsis'] == 1]['type'].value_counts())

Edge Cases and Validation¶

Handled Edge Cases¶

Blood culture times: Uses collect_dttm (collection time) as primary timestamp
Vancomycin routes: Treats IV and oral vancomycin as different antibiotics
Procedural medications: Excludes vasopressors given in procedural locations
Lab outliers: Caps extreme values (Cr ≤20, Bili ≤80, Plt ≤2000, Lactate ≤30)
Platelet baselines: Only uses baseline if any value ≥100 exists
ESRD exclusion: Patients with ESRD ICD codes excluded from AKI criterion