Heart-rate-based clock alignment in intensive-care ECGs, serial-ECG matching and administrative follow-up in MIMIC-IV

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Intensive care patients undergoing routine electrocardiography (n=129259). Intervention: Heart-rate-based ECG clock alignment and serial-ECG matching audit vs. Intermittently charted ICU heart rates and null permutation/shift constructions. Primary outcome: Identifiable individual ECG clock offsets meeting algorithmic criteria.

Leo Probst

doi:10.21203/rs.3.rs-11410281/v1

October 20, 2026Open Access

Heart-rate-based clock alignment in intensive-care ECGs, serial-ECG matching and administrative follow-up in MIMIC-IV

View Full Paper

Ask AI

Bookmark

Share

Key result

Heart-rate-based clock alignment identifies individual ECG offsets in only ~2% of ICU patients.

n=129,259

Why the study?

The study aimed to audit the timing, matching, and observation requirements of serial ECG studies that use routine clinical records in an intensive-care subset.

Key Points

To audit the timing, matching, and observation requirements of serial electrocardiogram studies that use routine clinical records in an intensive-care subset of MIMIC-IV.
Analyzed two person-disjoint development partitions of MIMIC-IV-ECG linked to MIMIC-IV comprising 17,067 and 5,846 ICU-linked individuals.
Evaluated heart-rate alignment against identifiability criteria, simulated offsets, and circular-shift and within-person permutation null constructions, alongside serial-matching calipers and 365-day administrative follow-up spans.
Individual clock offsets were identifiable for only 392 of 17,067 (2.3%) and 145 of 5,846 (2.5%) ICU-linked individuals, with 91.3% and 91.5% lost prior to estimation due to ICU stay boundaries and reference gaps.
Among identifiable offsets, 67.1% (95% CI 62.2–71.6%) and 70.3% (95% CI 62.4–77.3%) were within ± 60 min, compared to 86% achieved under zero-offset simulation.
Widening baseline potassium tolerance from 0.3 to 1.0 mmol/L expanded matched sets from 352 and 69 to 562 and 136, while 34.6% and about 36% of index discharges lacked any subsequent ECG or hospital discharge.

Study Design

Type

Observational (n=129,259)

Multicenter

No

Structured PICO

P

Population

129,259 patients (including 22,913 ICU-linked patients) with 640,204 routine ECG recordings at Beth Israel Deaconess Medical Center between 2008 and 2019, audited for clock alignment and up to 365-day administrative follow-up.

E

Exposure

Heart-rate-based clock alignment (matching ECG machine-reported mean RR heart rate to intermittently charted ICU heart rate across candidate offsets from -12 h to +12 h in 5 min steps), serial-ECG matching feasibility screening (potassium and calcium/heart-rate pairs), and administrative follow-up span auditing.

C

Comparator

Simulated known offsets (0, ±30, ±60, ±120, and ±240 min) and two null constructions that weaken alignment: within-person permutation of ECG rates (200 replications) and circular shifts of each stay's chart series (500 replications).

O

Outcome

Identifiability of individual ECG device-clock offsets (requiring ≥3 complete ECGs, connected near-optimal region within 1 bpm of minimum no wider than 120 min, minimum error ≤5 bpm, no near-optimal point at boundary, median offset within ±15 min, and ≥90% of identifiable offsets within ±60 min).

In routine ICU electronic health records from MIMIC-IV, individual ECG device-clock offsets could rarely be identified via heart-rate matching against charted rates, serial-ECG sample sizes depended heavily on matching tolerances, and administrative follow-up had substantial loss of contact within one year.

Limitations

Only ECGs nominally inside ICU stays were used for the alignment audit
Complete ±12 hour offset coverage was required, causing 91% structural loss before estimation
Heart rate was documented intermittently (mostly hourly) and reference origin (monitor vs manual entry) was uncoded
No calibrated individual uncertainty interval for the clock offset was produced
All MIMIC results used previously inspected development partitions without independent held-out confirmation
All MIMIC results use person-disjoint, previously inspected development partitions with no independent held-out confirmation reported.
91% of ICU-linked people lacked three ECGs with complete ±12 h reference coverage, largely due to ECG proximity to ICU stay boundaries (48% to 49%).
Analyses added after initial inspection (simulations, null constructions, individual support, reference restrictions, loss decomposition, exact-timestamp count, MC-MED diagnosis) were post hoc and the work was not prospectively registered.
Whether a charted heart rate value came from a monitor or was typed in is not coded, so machine RR and charted rate are not independent measurements and documentation coupling cannot be excluded.
Administrative follow-up contact cessation marks loss of visible documentation rather than verified event freedom, and emergency-department stays and laboratory events were not counted as contact.

Implication

Clock offsets are rarely identifiable via heart-rate matching in routine ICU ECGs; leaves open whether direct timestamps or advanced methods can enhance serial matching.

Abstract

Objective: To audit the timing, matching and observation requirements of serial electrocardiogram (ECG) studies that use routine clinical records in an intensive-care subset (9% of recordings). Approach: We used two person-disjoint, previously inspected development partitions of MIMIC-IV-ECG linked to MIMIC-IV. Intensive care unit (ICU) heart-rate alignment was assessed against identifiability criteria, simulated offsets and two null constructions. Pairs: development, internal assessment. Main results: Offsets were identifiable for 392 of 17,067 (2.3%) and 145 of 5,846 (2.5%) ICU-linked people; 91% were lost before estimation. Among people with at least three complete ECGs, the criteria were met in 27.1% as observed and in 19% to 22% under null constructions that weaken the alignment. Only 0.5% of ICU-linked people (117 of 1,979, 5.9%) had an offset with additional algorithmic support, not a calibrated uncertainty. Of the identifiable, 67.1% and 70.3% lay within ± 60 min (86% in simulation at zero offset), so the missed 90% criterion tested the procedure, not the clocks. Potassium matching limits rose from 352 and 69 to 562 and 136 sets when the baseline tolerance widened from 0.3 to 1.0 mmol/L. With exact timestamps, 34.6% and about 36% of index discharges had no later ECG or hospital discharge (36.7% and about 38% including deaths dated on or before discharge; emergency visits not counted as contact), and the follow-up span ended before 365 days in 26.4% and 25.6% of people with later contact and no documented atrial fibrillation or flutter (AF). A control in MC-MED (emergency department), locked before analysis and not a validation, missed its prespecified threshold (28.3% of 11,913 eligible visits), neither confirming nor refuting the MIMIC results. Significance: Individual clock offsets were rarely identifiable by heart-rate matching against intermittently charted rates. Matching counts depend on tolerances, and absent later contact is not absent events.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Leo Probst (2026) conducted an observational in Intensive care patients undergoing routine electrocardiography (n=129,259). Heart-rate-based ECG clock alignment and serial-ECG matching audit vs. Intermittently charted ICU heart rates and null permutation/shift constructions was evaluated on Identifiable individual ECG clock offsets meeting algorithmic criteria. Heart-rate-based clock alignment identified individual ECG offsets in only 2.3% to 2.5% of ICU-linked patients, with just 0.5% showing additional algorithmic support.

synapsesocial.com/papers/6ae28f3dcbadf70daf6c4445 https://doi.org/10.21203/rs.3.rs-11410281/v1

View Full Paper

Ask AI

Bookmark

Share

Heart-rate-based clock alignment in intensive-care ECGs, serial-ECG matching and administrative follow-up in MIMIC-IV

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Heart-rate-based clock alignment in intensive-care ECGs, serial-ECG matching and administrative follow-up in MIMIC-IV

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading