hannesill/m4/first-icu-stay

First ICU Stay Selection

Identifies first ICU stays and hospital admissions for cohort construction. Critical for avoiding correlated observations and inflated sample sizes from readmissions.

When to Use This Skill

• Creating research cohorts with independent observations
• Excluding readmissions that may have different characteristics
• Selecting incident (first-time) ICU patients
• Matching cohorts by admission characteristics

Pre-computed Table

SELECT
    subject_id,
    hadm_id,
    stay_id,
    -- Demographics
    gender,
    dod,  -- Date of death (if applicable)
    -- Hospital admission
    admittime,
    dischtime,
    los_hospital,  -- Hospital length of stay (days)
    admission_age,
    race,
    hospital_expire_flag,
    hospstay_seq,       -- Hospital admission sequence number
    first_hosp_stay,    -- TRUE if first hospitalization
    -- ICU stay
    icu_intime,
    icu_outtime,
    los_icu,            -- ICU length of stay (days)
    icustay_seq,        -- ICU stay sequence within hospitalization
    first_icu_stay      -- TRUE if first ICU stay of this hospitalization
FROM mimiciv_derived.icustay_detail;

Key Columns Explained

Hospital-Level

• hospstay_seq: Order of hospital admission for this patient (1 = first ever)
• first_hosp_stay: TRUE if this is the patient's first hospital admission

ICU-Level (Within Hospitalization)

• icustay_seq: Order of ICU stays within this hospitalization (1 = first ICU stay)
• first_icu_stay: TRUE if this is the first ICU stay of the current hospitalization

Age Calculation

Age is calculated as:

anchor_age + (YEAR(admittime) - anchor_year)

Important: MIMIC-IV shifts ages for patients > 89 years old at any admission. These patients have anchor_age = 91 and ages are not exact.

Common Cohort Selection Patterns

First ICU Stay Only (Most Common)

SELECT *
FROM mimiciv_derived.icustay_detail
WHERE first_icu_stay = TRUE;

First-Ever ICU Stay (Strictest)

SELECT *
FROM mimiciv_derived.icustay_detail
WHERE first_hosp_stay = TRUE
    AND first_icu_stay = TRUE;

Adult Patients Only

SELECT *
FROM mimiciv_derived.icustay_detail
WHERE admission_age >= 18
    AND first_icu_stay = TRUE;

Exclude Very Short Stays

SELECT *
FROM mimiciv_derived.icustay_detail
WHERE first_icu_stay = TRUE
    AND los_icu >= 1  -- At least 24 hours in ICU
    AND los_hospital >= 1;

Critical Implementation Notes

1. Multiple ICU Stays: One hospitalization can have multiple ICU stays. first_icu_stay identifies the first ICU stay within each hospitalization.

1. Patient-Level vs Stay-Level: Choose the right level of independence:

• first_hosp_stay = TRUE AND first_icu_stay = TRUE: Most independent
• first_icu_stay = TRUE: Independent within each hospitalization
• All stays: May have correlated outcomes

1. Age Shifting: Ages > 89 are shifted to protect privacy. Anchor age is 91, actual age is approximated.

1. Mortality Flags:

• hospital_expire_flag: Died during this hospitalization
• dod: Date of death (may be after discharge)

1. LOS Calculations:

• los_hospital: Days from admission to discharge
• los_icu: Days from ICU admission to ICU discharge

Example: Standard Adult Cohort

SELECT
    stay_id,
    subject_id,
    hadm_id,
    admission_age,
    gender,
    los_icu,
    hospital_expire_flag
FROM mimiciv_derived.icustay_detail
WHERE first_icu_stay = TRUE
    AND admission_age >= 18
    AND admission_age < 90  -- Exclude age-shifted patients
    AND los_icu >= 1;

Example: Readmission Analysis

-- Patients with multiple ICU stays in same hospitalization
SELECT
    hadm_id,
    COUNT(*) AS n_icu_stays
FROM mimiciv_derived.icustay_detail
GROUP BY hadm_id
HAVING COUNT(*) > 1;
 
-- Compare first vs subsequent stays
SELECT
    first_icu_stay,
    COUNT(*) AS n_stays,
    ROUND(AVG(los_icu), 1) AS avg_los,
    ROUND(AVG(hospital_expire_flag), 3) AS mortality
FROM mimiciv_derived.icustay_detail
GROUP BY first_icu_stay;

Example: Age Distribution

SELECT
    CASE
        WHEN admission_age < 40 THEN '<40'
        WHEN admission_age < 60 THEN '40-59'
        WHEN admission_age < 80 THEN '60-79'
        WHEN admission_age < 90 THEN '80-89'
        ELSE '90+'
    END AS age_group,
    COUNT(*) AS n_patients
FROM mimiciv_derived.icustay_detail
WHERE first_icu_stay = TRUE
GROUP BY 1
ORDER BY 1;

Related Exclusion Criteria

Common additional exclusions for research cohorts:

• Comfort care / DNR orders if your local dataset includes a validated code-status concept
• Organ donors
• Very short stays (< 4-24 hours)
• Missing key data elements
• Specific diagnoses (trauma, cardiac surgery, etc.)

References

• Johnson AEW et al. "MIMIC-IV, a freely accessible electronic health record dataset." Scientific Data. 2023.