Real-World Evidence Study Protocol (v2.0): Early Arterial Catheter Use and 28-Day Mortality in Mechanically Ventilated ICU Patients (MIMIC-III)

Retrospective ICU EHR cohort using explicit time zero (MV start) and early exposure window for IAC; pragmatic covariate strategy under data-access constraints

1. Title page

Element	Details
Title	Early Indwelling Arterial Catheter Use and 28-Day Mortality in Mechanically Ventilated ICU Patients: A Retrospective Cohort Study Using MIMIC-III EHR Data (Protocol v2.0).
Primary Objective	Estimate the adjusted association between early IAC exposure and 28-day mortality among mechanically ventilated adult ICU patients not receiving vasopressors in the first 24h after MV start.
Secondary Objectives	(Planned, contingent on data feasibility) Assess whether early IAC exposure is associated with (i) ICU length of stay (LOS) and (ii) monitoring intensity (e.g., ABG/day) if event tables are available.
Exposure Definition	Early IAC use: evidence of arterial line placement/documentation occurring within 0–24 hours after MV start (time zero). Exposure is derived from procedure/event documentation in MIMIC-III (not aline_flg).
Protocol Version	v2.0 (18-Dec-2025) (Active)
Contributors	PI: Ousmane Diallo, MPH-PhD (Design, Analysis, Reporting) Contact: ousmanerabi12@gmail.com
Study Registration	Not registered (portfolio RWE training project). Registration may be added later (OSF/ClinicalTrials.gov: TBD).
Sponsor	Self-sponsored project for RWE training and portfolio development.
Conflict of Interest	None declared.
Funding	Self-funded; no industry or external support.

Protocol status: This is the active protocol used for cohort construction and analysis. Protocol v1.0 (MIMIC-II) is archived for transparency and traceability.

2. Abstract

Background

Indwelling arterial catheters (IACs) are widely used in intensive care units (ICUs) for continuous blood pressure monitoring and frequent arterial blood gas (ABG) sampling, especially among mechanically ventilated (MV) patients. While IACs may improve monitoring, they are invasive and can be associated with complications (e.g., vascular injury, infection). Observational associations between IAC use and mortality are difficult to interpret because sicker patients are more likely to receive an IAC (confounding by indication) and because exposure timing can introduce immortal time bias if not handled carefully.

Research question and objectives

Among mechanically ventilated adult ICU patients who are not receiving vasopressors in the first 24 hours after MV initiation, is early IAC exposure (0–24h after MV start), compared with no early IAC exposure, associated with:

• primary outocome : 28-day all-cause mortality

• Secondary outcomes : ICU length of stay and monitoring intensity (e.g., ABG/day) (feasibility-dependent)

Methods

We will conduct a retrospective, observational analysis using the MIMIC-III ICU EHR database. We define time zero as the earliest timestamp indicating invasive mechanical ventilation during an ICU stay (MV start). We classify patients as exposed if an arterial line is documented/placed within 0–24 hours after MV start. Patients without documented arterial line placement during the exposure window are classified as unexposed.

To reduce confounding, we will estimate adjusted associations using

• 1. multivariable logistic regression and
• 2. propensity score methods (matching or weighting).

Due to computational/storage constraints, high-frequency event tables (e.g., chartevents, labevents) may be unavailable; therefore, baseline covariate adjustment will rely on variables feasible to extract (e.g., demographics, admission characteristics, comorbidities/diagnoses, ICU/service type, and vasoactive medication exposure from input tables). Results will be reported as odds ratios with 95% confidence intervals, with transparent reporting of limitations due to unavailable physiologic/laboratory covariates.

3. Amendments and updates

Version date	Version number	Section amended	Amendment / update	Reason for amendment
27-Nov-2025	v1.0	Full protocol	Initial draft (MIMIC-II framing, aline_flg exposure).	Initial specification under presumed data constraints.
18-Dec-2025	v2.0	Data source, exposure, time zero, covariates, feasibility	Updated to MIMIC-III; reconstructed MV start as explicit time zero; defined early IAC exposure window (0–24h); implemented vaso-free day-1 criterion; adopted pragmatic covariate strategy excluding high-frequency event tables when unavailable.	Improve temporal validity and reduce immortal time bias; ensure feasibility and reproducibility given storage/compute limitations.
18-Dec-2025	v2.0	Population – Exclusion criteria	Excluded patients admitted to the Cardiac Surgery Recovery Unit (CSRU) (n = 60) due to systematic pre-ICU arterial line placement	Prevents structural exposure misclassification and improves comparability between exposed and unexposed groups

Note: All analyses in this project are based on Protocol v2.0.

4. Milestones

Milestone	Date
Protocol v2 finalization	December 2025
Cohort extraction (SQL)	December 2025
Descriptive analysis & QA	December 2025
PS model & balance diagnostics	January – February 2026
Primary analysis (mortality)	February 2026
Secondary analyses (if feasible)	February 2026
Sensitivity analyses	March 2026
Portfolio write-up	March 2026

5. Rationale and background

Mechanically ventilated ICU patients require close monitoring of ventilation, oxygenation, and hemodynamics. IACs can facilitate continuous invasive blood pressure monitoring and repeated ABG sampling, but their risks and benefits are debated. Observational ICU studies of IAC use are vulnerable to:

• Confounding by indication: clinicians preferentially place IACs in more severely ill patients.
• Immortal time bias: if exposure is defined as “ever got IAC” during a follow-up period beginning at MV start, exposed patients must survive long enough to receive an IAC. This can bias comparisons if not handled with an explicit exposure window aligned to time zero.

This protocol uses an explicit time zero (MV initiation) and a prespecified early exposure window (0–24 hours) to align patient eligibility, exposure classification, and follow-up, supporting clearer causal interpretation under real-world constraints.

6. Research question and objectives

A. Primary research question and objective

Study element	Specification
Objective	Estimate the association between early IAC exposure (0–24h after MV start) and 28-day mortality among mechanically ventilated adult ICU patients not receiving vasopressors in the first 24h after MV start.
Hypothesis	Early IAC exposure is not associated with lower 28-day mortality after adjustment (null hypothesis).
Population	Adult ICU patients in MIMIC-III who (1) received invasive MV, (2) meet eligibility/exclusion criteria, and (3) can be classified for exposure and outcome.
Exposure	Early IAC: arterial line documented/placed within 0–24h after MV start.
Comparator	No arterial line documentation/placement during the 0–24h exposure window (primary definition).
Outcome	28-day all-cause mortality (in-hospital death within 28 days of MV start, as available in MIMIC-III).
Time	Time zero: MV start. Follow-up: up to 28 days after MV start, or until hospital discharge/death (whichever occurs first, depending on available timestamps).
Setting	Adult ICUs captured in MIMIC-III (Beth Israel Deaconess Medical Center).
Main measure of effect	Adjusted odds ratio (aOR) for 28-day mortality from multivariable logistic regression and propensity-score–based models.

B. Secondary research questions and objectives (feasibility-dependent)

Study element	Specification
Objective	Assess whether early IAC exposure is associated with (1) ICU LOS, and (2) monitoring intensity (e.g., ABG/day) when required tables are available.
Population	Same as primary analysis.
Exposure/Comparator	Same as primary analysis.
Outcomes	(1) ICU LOS in days; (2) ABG/day (requires ABG identification from event tables).
Time	For ICU LOS: entire ICU stay. For ABG/day: ICU stay (or MV period), depending on data availability.
Main measure of effect	LOS: adjusted mean difference or time-to-discharge model; ABG/day: Poisson/negative binomial rate ratio.

7. Research methods

Study design

Retrospective observational cohort study using routinely collected ICU EHR data from MIMIC-III.

This protocol is target-trial–inspired:

• Eligibility is assessed at/near time zero (MV start).
• Exposure is assigned within a prespecified window (0–24h after MV start).
• Follow-up begins at time zero for the outcome of interest.

This design helps reduce immortal time bias relative to an “ever exposed” definition.

Key design definitions

Time zero

MV initiation (MV start) within an ICU stay.

Exposure window (0–24h)

A patient is classified as early IAC exposed if the earliest documented arterial line placement/event occurs in [MV start, MV start + 24h).

Patients without arterial line evidence in this window are classified as unexposed (primary definition).

Important distinction
Time zero defines when follow-up begins (MV start).
The exposure window defines when exposure status is determined (0–24h after time zero).
These are not the same concept.

Settings and variables

Setting
ICU encounters from MIMIC-III (BIDMC). Data are queried from a PostgreSQL database (AWS RDS).

Primary exposure
- Early IAC placement/documentation within 0–24h of MV start (derived from procedure/event documentation).

Primary outcome
- 28-day all-cause mortality (in-hospital death within 28 days after MV start, based on available death time fields).

Key covariates (pragmatic, feasibility-driven)

Because high-frequency event tables (e.g., chartevents, labevents) may be unavailable due to storage constraints, baseline adjustment will prioritize variables extractable from smaller/feasible tables:

• Demographics: age (at ICU admission), sex, ethnicity (from admissions/demographics tables as available)
• Admission characteristics: admission type, service category (medical vs surgical), ICU stay order
• Comorbidity proxies: ICD-9 diagnosis-based comorbidity measures (e.g., Charlson) where feasible
• Vasopressor exposure: vasopressor administration in first 24h after MV start (used as an exclusion and/or covariate)

Explicitly not included in v2.0 (due to data constraints) - Day-1 labs and high-resolution vitals from labevents/chartevents, if these tables are unavailable in the working environment.

Inclusion/exclusion criteria (operational)

1. Adult patients: age ≥ 16 years
2. Mechanically ventilated during ICU stay; MV start defined
3. MV start occurs early relative to ICU admission: within first 24h of ICU admission
4. ICU LOS above minimum threshold
5. First ICU stay per patient (to avoid correlated observations and simplify interpretation)
6. Exclude surgical service admissions
7. Exclude vasopressor use in the first 24h after MV start (primary population definition)
8. Patients admitted to the Cardiac Surgery Recovery Unit (CSRU) were excluded (n = 60). In this unit, indwelling arterial catheters are frequently placed prior to ICU admission and prior to initiation of mechanical ventilation, leading to structural exposure misclassification and lack of comparability with non-exposed patients.

Statistical analysis plan

Descriptive analysis

• Summarize baseline characteristics by early IAC exposure status.
• Report standardized mean differences (SMD) for covariate balance.

Primary analysis (28-day mortality)

• Multivariable logistic regression:
  • Outcome: death_28d
  • Predictor: early IAC exposure
  • Adjust for feasible confounders (demographics, admission/service features, comorbidity proxies)

Propensity score analysis

• Fit a propensity score model for early IAC exposure using the same feasible covariates.
• Apply:
  • 1:1 nearest neighbor matching or
  • IPTW with stabilized weights
• Re-estimate association with weighted/matched outcome model.
• Assess balance using SMD before/after.

Sensitivity analyses (planned)

• Alternative exposure definitions:
  • “Any IAC during ICU stay” (risk of immortal time bias; reported as sensitivity only)
  • Alternative exposure windows (e.g., 0–12h, 0–48h)
• Alternative eligibility rules:
  • Different LOS thresholds
  • Include vs exclude surgical service (covariate vs exclusion)
• Outcome definition sensitivity:
  • Use additional death fields (e.g., patient-level date of death) when available/appropriate and consistent with MIMIC-III.

Data management & reproducibility

• SQL-first cohort extraction; R used for analysis.
• Version control via GitHub; all derivations documented.
• Deliverables include cohort flowchart, SQL scripts, and analysis scripts.

Quality control

• Unit checks and sanity checks for timestamps (MV start within ICU stay bounds).
• Internal checks for duplicates (icustay_id uniqueness).
• Validation of exposure/outcome construction on random samples.

8. Limitations

• Residual confounding remains possible, particularly if severity-of-illness measures and high-resolution vitals/labs are unavailable.
• Exposure misclassification is possible because IAC documentation may be incomplete or recorded after placement.
• Single-center ICU data may limit generalizability.
• Operational definitions (e.g., MV start derived from available tables) may differ from bedside reality; all logic is explicitly documented.

9. Protection of human subjects

This project uses de-identified MIMIC-III data under the PhysioNet data use agreement. No re-identification is attempted. The project qualifies as non-human subjects research as conducted on de-identified data (institutional determinations may vary).

10. Reporting of adverse events

Not applicable (retrospective analysis of de-identified data).

11. References

1. Hernán MA, Robins JM. “Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available.” NEJM (2016).
2. Goldberger AL et al. “PhysioNet: Components of a New Research Resource for Complex Physiologic Signals.” Circulation (2000).
3. Johnson AEW et al. “MIMIC-III, a freely accessible critical care database.” Sci Data (2016).
4. Hsu DJ et al. (as previously cited; ICU arterial catheter observational context).

12. Appendices

• Appendix A: Cohort flowchart & eligibility logic (SQL-aligned)
• Appendix B: Variable definitions (time zero, exposure window, outcome)
• Appendix C: SQL queries and preprocessing workflow (PostgreSQL)
• Appendix D: Propensity score diagnostics (SMD/Love plot)
• Appendix E: Regression model specifications

Ousmane Diallo, MPH-PhD – Biostatistician, Data Scientist & Epidemiologist based in Chicago, Illinois, USA. Specializing in SAS programming, CDISC standards, and real-world evidence for clinical research.