BGPT: Paper Review: Risk factors of posthemorrhagic seizure in spontaneous intracerebral hemorrhage

Fuel Your Discoveries

Quick Explanation Copied

Core finding (from the paper text provided)

In 400 adults with spontaneous intracerebral hemorrhage (SICH), seizure occurrence (overall 17.5%) was independently associated with lobar hematoma location and hematoma volume > 10 mL, with GCS ≤ 12 additionally associated with early seizures, and midline shift additionally associated with late seizures; seizures were also linked to worse 2-year functional outcomes (mRS, spastic hemiparesis) in the study’s retrospective dataset.

Evidence strength: association only (retrospective, single-center, seizure ascertainment without routine EEG), so mechanisms and causality remain uncertain.

Long Explanation

Paper Review (Visual + Critical): Risk factors of posthemorrhagic seizure in spontaneous intracerebral hemorrhage

Evidence dates/metadata: received Sept 30, 2024; accepted Jan 15, 2025; published Jan 23, 2025.

Design: Retrospective cohort N=400 SICH adults Outcome: mRS @2 years

1) Patient selection + seizure timing buckets

Cohort flow (as described in the provided paper text)

Initial ICD-10 pool: 1,539 patients; exclusions left a final analytic cohort of 400 SICH adults.
Seizure occurrence: 70/400 (17.5%), split into early 30/400 (7.5%) and late 40/400 (10%); early and late seizure categories did not overlap in the study’s classification scheme.

2) Visual summary of seizure prevalence

Seizure bucket definitions (early vs late) follow the paper’s classification; this review does not infer a specific time threshold because the provided excerpt does not supply it explicitly.

3) Multivariable predictors: forest-style plot (odds ratios)

Adjusted ORs reported by the paper

Below, only variables with adjusted ORs explicitly provided in the paper text are visualized.

Source for ORs/CI/p-values: paper’s multivariable results section and Table 4.

4) Univariate-to-multivariable interpretation (what changed?)

The paper explicitly notes univariate associations for multiple clinical/radiographic factors, but only a subset remains statistically independent in multivariable models.

5) Strengths, then skeptical critique (biological/measurement validity)

What the paper did well

Seizures stratified by timing (early vs late) and analyzed with separate multivariable models (early, late, overall).
Radiographic lesion burden (lobar location, hematoma volume, midline shift) features prominently in adjusted models, aligning with mechanistic plausibility that cortical involvement and mass effect are epileptogenic. (Mechanistic plausibility is not proof of mechanism in humans.)

Critical limitations / likely sources of bias or uncertainty

Retrospective design: information bias and residual confounding are expected; the paper itself flags retrospective, single-center, and unmeasured-variable concerns.
Seizure ascertainment may be incomplete: the paper reports seizure evaluation from interviews at follow-up without routine EEG confirmation, which can under-detect electrographic or subtle seizures.
Treatment-indication confounding risk: “craniotomy with hematoma evacuation” appears as an adjusted predictor of overall seizures, but surgery is not randomly assigned—patients receiving surgery likely differed systematically (e.g., larger lesions/more severe presentations), even if some radiographic severity variables were adjusted.
Outcome linkage (seizure ↔ worse 2-year function) is associative: because seizures may coincide with markers of lesion burden and severity, the observed functional disadvantage may partly reflect baseline injury severity rather than seizure causal effects. The excerpted text states association with higher mRS/spastic hemiparesis at 2 years, but causality is not established.

6) External context checks (does it agree with other evidence?)

Convergent themes across related literature (selection only where DOIs were provided)

BBB/vascular integrity disruption and seizures are mechanistically linked in experimental work: endothelial β-catenin signaling maintains adult BBB integrity; its disruption in adult mice leads to BBB leakage, CNS inflammation, seizures, and petechial hemorrhages. This supports biological plausibility for hemorrhage-triggered epileptogenesis pathways, though the present review paper is observational.
Seizures after ICH are repeatedly associated with worse outcomes in large observational registries (example provided here), though measurement differences (especially EEG confirmation) can alter incidence estimates and weaken causal interpretation.
Predictors of seizure onset differ across cohorts and are sensitive to inclusion criteria and seizure definitions: a surgical-ICH cohort study reported predictors such as volume and SAH/SDH for early seizures and INR/SDH for late seizures. This supports the idea that “risk factor sets” may be cohort- and protocol-dependent.
Seizure prophylaxis effects are debated and confounded by indication: observational comparisons show reduced in-hospital seizure incidence with newer AED prophylaxis but uncertain improvements in functional outcomes. This underscores that even if risk factors are identifiable, intervention effects cannot be inferred from observational seizure-risk studies alone.

7) Practical takeaways (stated strictly as associations, not causation)

High-risk clinical-radiographic phenotype for seizure occurrence in this dataset: lobar hematoma and hematoma volume >10 mL (overall seizures); for timing-specific risk: GCS ≤12 for early seizures, and midline shift for late seizures.
Seizure occurrence correlated with worse 2-year functional outcomes in the paper’s follow-up assessment (spastic hemiparesis and unfavorable mRS); however, because this is observational, the directionality/causality remains uncertain.

8) What evidence would most strongly disprove/redirect these conclusions?

A prospective multi-center study with routine standardized EEG adjudication and precise seizure onset timing could test whether lobar location and hematoma volume remain independent predictors when electrographic seizures are not missed.
Analyses that properly handle confounding by indication for surgery could test whether the adjusted association of craniotomy/evacuation with overall seizures persists.
If seizure detection improves while controlling severity and lesion burden, and seizure groups no longer show worse mRS after adjudication, the observed seizure–outcome relationship could be largely explained by baseline severity rather than seizure effects.

9) Author review shortcuts

Open focused author-centric pages on BGPT.

Feedback:

Updated: April 16, 2026