apsiii-score

name	apsiii-score
description	Calculate APACHE III (Acute Physiology Score III) for ICU patients in MIMIC-IV. Use for mortality prediction, severity stratification, case-mix adjustment, or risk-adjusted outcome comparisons.
license	Apache-2.0
metadata	{"author":"m4-clinical-extraction","version":"1.0","database":"mimic-iv","category":"severity-scores","source":"https://github.com/MIT-LCP/mimic-code/tree/main/mimic-iv/concepts/score","validated":true}

APACHE III (APS III) Score Calculation

The Acute Physiology Score III (APS III) is the physiological component of APACHE III. It measures patient severity of illness on the first day of ICU stay and provides hospital mortality probability estimates.

When to Use This Skill

• Hospital mortality prediction
• Severity stratification and risk adjustment
• Case-mix adjustment for benchmarking
• Comparing outcomes across ICUs or time periods
• Research cohort severity matching

Score Components

The APS III uses the worst values from the first 24 hours of ICU stay:

Vital Signs

• Heart rate (normal reference: 75 bpm)
• Mean blood pressure (normal reference: 90 mmHg)
• Temperature (normal reference: 38C)
• Respiratory rate (normal reference: 19 breaths/min)

Laboratory Values

• Hematocrit (normal reference: 45.5%)
• WBC (normal reference: 11.5 x10^9/L)
• Creatinine (normal reference: 1.0 mg/dL)
• BUN (scored from 0, higher is worse)
• Sodium (normal reference: 145.5 mEq/L)
• Albumin (normal reference: 3.5 g/dL)
• Bilirubin (scored from 0, higher is worse)
• Glucose (normal reference: 130 mg/dL)

Blood Gas

• PaO2 (for non-ventilated patients with FiO2 < 50%)
• A-aDO2 (for ventilated patients with FiO2 >= 50%)
• pH and PaCO2 interaction scoring

Other

• GCS (complex interaction between eye, verbal, motor components)
• Urine output (24-hour total)
• Mechanical ventilation status
• Acute renal failure flag

Pre-computed Table

SELECT
    subject_id,
    hadm_id,
    stay_id,
    apsiii,
    apsiii_prob,  -- Predicted hospital mortality probability
    hr_score,
    mbp_score,
    temp_score,
    resp_rate_score,
    pao2_aado2_score,
    hematocrit_score,
    wbc_score,
    creatinine_score,
    uo_score,
    bun_score,
    sodium_score,
    albumin_score,
    bilirubin_score,
    glucose_score,
    acidbase_score,
    gcs_score
FROM mimiciv_derived.apsiii;

Critical Implementation Notes

1. Worst Value Definition: "Worst" means furthest from a predefined normal reference value, not simply min or max. For example:

   • Heart rate worst = MAX(|HR - 75|)
   • If equally distant from normal, use the higher score

2. Acute Renal Failure (ARF) Modifier: ARF is defined as:

   • Creatinine >= 1.5 mg/dL AND
   • Urine output < 410 mL/day AND
   • No chronic kidney disease (CKD stages 4-6)

3. Ventilation Interaction:

   • For ventilated patients with FiO2 >= 50%: use A-aDO2
   • For non-ventilated patients with FiO2 < 50%: use PaO2
   • Only arterial blood gas specimens are used

4. pH/PaCO2 Interaction: The acid-base score requires both pH and PaCO2 together - different combinations yield different scores.

5. GCS Scoring: Complex interaction matrix between eye, verbal, and motor scores. Sedated/intubated patients default to normal (score 0).

6. Temperature: Axillary measurements should theoretically be increased by 1 degree, but this is not implemented.

7. Mortality Probability: Calculated using logistic regression:

   apsiii_prob = 1 / (1 + exp(-(-4.4360 + 0.04726 * apsiii)))
   

Example: Get Severity Distribution

SELECT
    CASE
        WHEN apsiii < 30 THEN 'Low (<30)'
        WHEN apsiii < 60 THEN 'Moderate (30-59)'
        WHEN apsiii < 90 THEN 'High (60-89)'
        ELSE 'Very High (>=90)'
    END AS severity_category,
    COUNT(*) AS n_patients,
    ROUND(AVG(apsiii_prob), 3) AS avg_predicted_mortality
FROM mimiciv_derived.apsiii
GROUP BY 1
ORDER BY 1;

Example: Compare Predicted vs Actual Mortality

SELECT
    ROUND(apsiii_prob, 1) AS predicted_mortality_decile,
    COUNT(*) AS n_patients,
    SUM(adm.hospital_expire_flag) AS actual_deaths,
    ROUND(AVG(adm.hospital_expire_flag), 3) AS observed_mortality
FROM mimiciv_derived.apsiii a
INNER JOIN mimiciv_hosp.admissions adm
    ON a.hadm_id = adm.hadm_id
GROUP BY 1
ORDER BY 1;

References

• Knaus WA et al. "The APACHE III prognostic system: Risk prediction of hospital mortality for critically ill hospitalized adults." Chest. 1991;100(6):1619-1636.
• Johnson AEW. "Mortality prediction and acuity assessment in critical care." University of Oxford. 2015. (Calibration equation source)