CMR atrial measures track AF readmission risk in small retrospective cohort

Cardiovascular magnetic resonance measurements of left atrial function were associated with 30-day unplanned readmission among patients hospitalised with atrial fibrillation in a small, single-centre retrospective study at Chulabhorn Hospital in Bangkok. In the 18 admitted patients included in the primary comparison, lower left atrial reservoir strain and a higher ratio of left atrial volume index to left ventricular ejection fraction showed strong but statistically uncertain discrimination between those readmitted and those who were not. The findings suggest a possible role for magnetic resonance imaging in identifying vulnerable patients, but they do not establish an independent effect, a causal mechanism or a clinically usable prediction rule.

The cohort, named CMRRADAR-AF, included 43 adults with atrial fibrillation whose cardiovascular magnetic resonance scans were acquired on 1.5-T or 3-T Philips systems. Patients were identified through electronic medical records from January 2018 to December 2024, and the study included 25 outpatients managed without hospitalisation, 10 patients admitted for an AF-related cause without unplanned readmission within 30 days, and eight patients readmitted within that period. The primary comparison therefore included 18 admitted patients and only eight readmission events. Overall, 53% of participants were men, mean age was 66 years, 63% had paroxysmal AF and 37% had long-standing persistent AF. Hypertension was the most common comorbidity, affecting 84% of the cohort.

Left atrial reservoir strain was derived by feature tracking from standard two- and four-chamber cine images. It measures the maximum longitudinal stretch of the atrial wall from the end of left ventricular diastole to the point of maximum left atrial volume during ventricular systole, while the mitral valve is closed. This phase does not require organised atrial contraction, making it measurable in AF. Booster-pump strain and conduit strain were not reported because their derivation requires identifying atrial contraction or mitral valve opening, which could not be done reliably during AF. Analysis was performed with CVI42 version 6.4. Left atrial volume was measured with the biplane area-length method and indexed to body-surface area, while ventricular volumes, ejection fractions and other structural measurements were derived from semiautomated contouring.

Among admitted patients, median left atrial reservoir strain was 6.6% in those readmitted and 15.4% in those who were not, a difference with a p-value below 0.001. The strain measure had an area under the receiver operating characteristic curve, or AUC, of 0.963, with a 95% confidence interval of 0.90 to 1.00. An exploratory threshold derived within the same dataset, below 11.1%, identified 88% of readmissions, with a 95% confidence interval of 47% to 100%, and classified every patient without readmission correctly, with a specificity of 100% and a 95% confidence interval of 69% to 100%. Because the threshold and predictor were selected in the same small sample, the investigators estimated an optimism-corrected AUC of 0.919 after accounting for selection. These figures describe discrimination within this exploratory cohort, not validated performance in an unselected AF population.

The ratio combining left atrial volume index with left ventricular ejection fraction also showed an association with readmission. Its median value was 1.21 in readmitted patients and 0.57 in those without readmission, and its AUC was 0.913, with a p-value of 0.002. At an exploratory threshold of at least 0.77, sensitivity was 100% and specificity was 80%. Left atrial volume index alone produced an AUC of 0.850, whereas right ventricular ejection fraction produced an AUC of 0.838. DeLong comparisons did not show significant differences between left atrial reservoir strain and the ratio, between the ratio and left atrial volume index alone, or between strain and volume index alone. With 18 subjects and eight events, however, these non-significant results are inconclusive rather than evidence that the measures perform equally well. The ratio was strongly correlated with left atrial volume index, which forms its numerator, so the study could not establish that dividing by ejection fraction added independent information.

Qualitative atrial hyperenhancement was observed on late gadolinium enhancement imaging in all eight readmitted patients and in none of the 10 admitted patients who were not readmitted. This complete separation was striking but was not converted into a formal discrimination statistic. The scans used a late gadolinium enhancement protocol optimised for nulling left ventricular myocardium rather than for characterising the much thinner atrial wall, which averages 2 to 3 mm. Dedicated atrial protocols use shorter inversion times, higher spatial resolution and respiratory-gated three-dimensional acquisition. The study’s binary assessment therefore cannot be treated as validated quantification of atrial fibrosis, and the authors did not interpret the pattern as proof of a fibrosis mechanism. Non-ischaemic left ventricular scar was present in five of eight readmitted patients and in none of the non-readmitted admitted patients, a descriptive association that was not used to establish an atrial substrate mechanism.

Readmission causes were heterogeneous. Five of the eight readmissions involved AF with rapid ventricular response and three involved bleeding associated with anticoagulation. In a post hoc analysis excluding the bleeding events, left atrial reservoir strain retained an AUC of 0.940, indicating that the overall association was not driven solely by bleeding. At the same time, the three patients readmitted for bleeding also had low strain, with a median of 7.5%, suggesting that low reservoir strain may identify broadly higher-risk patients rather than a group with specifically arrhythmic recurrence. A prespecified sensitivity analysis restricted to 14 patients whose scans had been obtained within 12 months of the index admission, including six readmissions, produced an AUC of 0.938 for reservoir strain. The median scan-to-admission interval for the full admitted group was 134 days, although the interval was longer in the readmitted group than in the non-readmitted group: 280 versus 78 days, a difference that did not reach statistical significance.

The three-group comparison revealed a non-monotonic pattern. Patients managed as outpatients had a median reservoir strain of 8.5%, lower than the 15.4% recorded among admitted patients who were not readmitted, while the readmitted group had the lowest value at 6.6%. The corresponding median ratios of left atrial volume index to left ventricular ejection fraction were 0.83, 0.57 and 1.21, respectively. Right ventricular ejection fraction and left atrial ejection fraction showed a comparable pattern, with the non-readmitted admitted group having the highest values. These administratively defined groups do not constitute data-derived phenotypes, and the pattern could reflect differences in underlying illness, referral practices or reversible causes of admission. Notably, heart-failure history was present in 75% of readmitted patients, compared with 20% of admitted patients who were not readmitted and 8% of outpatients, making it impossible to separate an AF-specific atrial effect from shared structural and functional disease. The investigators therefore described the three-group results as hypothesis-generating rather than evidence of a progressive disease continuum.

The LACE clinical risk score, which combines length of stay, admission acuity, Charlson comorbidity index and recent emergency-department visits, had an AUC of 0.719, with a 95% confidence interval of 0.44 to 0.95. Its median score was 11 in the readmitted group and 9.5 in the non-readmitted group, but the difference was not statistically significant, with p equal to 0.125. At the published high-risk threshold of 10 or more, sensitivity was 75% and specificity was 50%. An exploratory model combining LACE with reservoir strain produced a numerically higher AUC, yet the investigators did not claim incremental value because eight events were insufficient for a stable model. They also could not perform reliable multivariable adjustment: adding two predictors to this dataset would reduce the number of events per variable below accepted levels, while correlated imaging measures could further destabilise the estimates. The apparent performance of the imaging measures therefore cannot be interpreted as superiority over LACE or as proof that imaging adds clinically useful prediction.

Measurement reproducibility was high in a subset of 22 scans. For reservoir strain, the intraclass correlation coefficient was 0.998 for repeated measurements by the same reader and 0.952 between readers; the mean inter-observer absolute difference was 1.5 percentage points, with limits of agreement from minus 3.7 to plus 3.2 percentage points. Agreement was similarly high for left atrial volume index and left ventricular ejection fraction. Other major limitations included the retrospective design, selection of patients referred for magnetic resonance imaging, reliance on one hospital’s records, inability to capture readmissions elsewhere, and inability to reliably retrieve heart rate and rhythm at the time of imaging. Larger prospective, multicentre studies with uniform scan timing, dedicated atrial late gadolinium enhancement, direct comparisons with clinical models and adjustment for heart failure are needed before these measures can be used for discharge risk stratification or routine decision-making.

Subject of Research: Exploratory associations between cardiac MRI measures and 30-day readmission after atrial fibrillation hospitalisation

Article Title: CMR-derived left atrial reservoir strain and the LAVI/LVEF ratio are associated with 30-day readmission in atrial fibrillation: the CMRRADAR-AF retrospective cohort study

Article References: Panyawongkhanti, M., Fueangtalop, A., Satitthammaphon, N., Thaitirarot, C., Mueangsilpsat, V., Wuttichaipradit, C., Tunhasiriwer, A., & Tanking, C. (2026). CMR-derived left atrial reservoir strain and the LAVI/LVEF ratio are associated with 30-day readmission in atrial fibrillation: the CMRRADAR-AF retrospective cohort study. Frontiers in Cardiovascular Medicine, 13, Article 1845659. https://doi.org/10.3389/fcvm.2026.1845659

Image Credits: AI Generated

DOI: 10.3389/fcvm.2026.1845659

Keywords: atrial fibrillation, cardiac MRI, left atrial strain, readmission, LAVI/LVEF ratio, heart failure, risk stratification, late gadolinium enhancement, LACE score, Thailand, CMR-derived, left