Concerns in a National Inpatient Sample study of peripartum cardiovascular events in autoimmune rheumatic disease

Regmi and colleagues used the National Inpatient Sample (NIS) to examine peripartum cardiovascular events in delivery hospitalizations among women with autoimmune rheumatic diseases (ARDs).¹ The clinical question is important, and the directional findings for systemic lupus erythematosus are plausible. We raise two concerns about the vasculitis-subgroup results, neither addressed in the paper’s limitations: the headline odds ratios rest on very sparse data, and several figures in the abstract differ from the article’s own tables.

First, the most striking associations rest on extremely sparse data. The vasculitis cohort comprised 2578 hospitalizations, within which the reported events were few: 10 strokes (0.39%), 25 cases of peripartum cardiomyopathy (0.97%), 50 cases of acute kidney injury (1.94%), and 35 of heart failure (1.36%) (Table 4). The adjusted odds ratios derived from these counts are correspondingly unstable: stroke 31.9 (95% CI 4.15–246.3), acute kidney injury 67.4 (28.6–158.6), peripartum cardiomyopathy 28.4 (11.06–73.28), and heart failure 12.08 (4.36–33.5) (Table 5). Confidence intervals spanning one to two orders of magnitude are the signature of sparse-data bias, in which maximum-likelihood logistic estimates become inflated and unreliable.² That these intervals exclude unity does not establish a reliable association: sparse-data bias systematically inflates the estimate away from the null, so the statistical significance reflects this instability rather than refuting it. Several of the underlying cells are very small (stroke in the vasculitis group and peripartum cardiomyopathy in the rheumatoid arthritis group are each reported as only 10 events), at the small-cell threshold (10 or fewer) that the Healthcare Cost and Utilization Project treats as too few to report.³ These estimates convey little beyond "harm cannot be excluded" and should not be presented as point estimates of effect without bias-reducing methods (for example, penalized-likelihood regression) or an explicit statement of their instability; the absence of a described survey design (discharge weights, strata, and clusters) further limits their precision.⁴

Second, the abstract and the body of the paper report different numbers. The abstract gives a vasculitis stroke odds ratio of 513.3, peripartum cardiomyopathy 66.7, and heart failure 24.17, with vasculitis mortality of 1.54% versus 0.0001%; the corresponding values in the Results text and Tables 5 and 6 are 31.9, 28.4, 12.08, and 0.96% versus 0.007%. The discrepancies extend to the lupus estimates (for example, arrhythmia 3.17 in the abstract versus 2.95 in Table 5; venous thrombosis 8.4 versus 5.84; preeclampsia 1.5 versus 1.75). Because abstracts are indexed and cited far more widely than full texts, and are the version most likely to be incorporated into reviews and meta-analyses, these inconsistencies risk propagating effect sizes an order of magnitude larger than the authors’ own analysis supports. A published correction reconciling the abstract with the tables would be valuable.

None of this overturns the broad message that ARDs, and vasculitis in particular, may carry elevated peripartum cardiovascular risk. But the magnitude of that risk cannot be read from the present estimates, and the discrepant abstract figures should be corrected so that the literature does not inherit them.

—Brett D. Cohen, DO Department of Internal Medicine, St. Luke’s University Health Network, Bethlehem, Pennsylvania, USA brettdavidcohen@gmail.com
http://orcid.org/0009-0005-1645-6381

THE AUTHORS RESPOND

We appreciate the opportunity to clarify the concerns raised regarding both the vasculitis subgroup analysis and the discrepancies between the abstract and the final reported results and offer corrections. The separate corrections note can be found at doi:10.1080/08998280.2026.2706407.

—Anil Regmi, MBBS, MD Department of Internal Medicine, Parkview Health, Fort Wayne, Indiana, USA dr.anilregmi7@gmail.com