Stimulated by Chen et al 2026.[1]
IF – impact factor
US – Unitied States of America
TriNetX – EHR from 67 HCOs in the US Collaborative Network
EHR – electronic health records
HCO – health care organisation
NHIRD – national health insurance research database (Taiwan)
PSM – propensity score matching
WBC – white blood cell count
ALT – alanine aminotransferase
AST – aspartate aminotransferase
GFR – glomerular filtration rate
RR – risk ratio
HR – hazard ratio– key to acronyms
This is one of the first papers I sent out for peer review when I took over screening new submissions to our journal Acupuncture in Medicine (IF 2.7) from David a few months ago.
It is another one of the large retrospective cohort studies, and it comes from a group in Taiwan, but it uses data from the US (TriNetX) rather than from the NHIRD in Taiwan.
The TriNetX database has nearly 115 million anonymised patient records, and from these the current research group found just over 3 million with biliary disorders. Of these, just under 8k had received acupuncture following first diagnosis. After PSM they were left with 7930 in each cohort.
PSM included age, sex, race, and the presence of diseases of the liver, diabetes mellitus, overweight and obesity, diseases of other endocrine glands, neoplasms, and hypertension.
The primary outcome was cholecystectomy withing 5 years of the index date. Secondary outcomes were all laboratory markers: WBC; procalcitonin; total bilirubin; ALP; ALT or AST; and GFR.
The index dates were a little different in the 2 cohorts. In the acupuncture cohort it was the first date of co-occurrence of both a biliary disease diagnosis code and an acupuncture provision code in the record. In the matched control cohort, it was the date of the first biliary diagnostic code.
This introduces a degree of immortal time bias but also a degree of selection bias, since the acupuncture cohort had to have sufficiently mild disease that they did not have an immediate cholecystectomy before they could seek acupuncture. The latter bias is probably more important and could have been mitigated by using a landmark analysis ie setting the index dates at the same time (landmark) following diagnosis in each cohort (say 6 months).
Cholecystectomy rates were considerably lower in the acupuncture cohort (7.6% vs 13.5% – RR 0.564). Rates of abnormal laboratory tests were also significantly lower in the acupuncture cohort.
On reviewing this paper, I came across something called an E-value for the first time. The E-value is a sensitivity analysis metric that quantifies how strong an unmeasured cofounder would need to be to explain away the observed association.
The E-value can be calculated from either an RR or an HR. In this case the authors used the HR (0.525), which led to an E-value of 3.22, which is considered large. If they had used the RR, the E-value would have come out as 2.94. Whilst both values are large, this does not address the problem of exposure-selection bias and immortal time bias.
The authors have assured me that they are aiming to address these and other concerns raised during peer review in their ongoing research.
References
1 Zhu F, Wang Y, Liu J, et al. Comparative effectiveness of different acupuncture courses for chronic insomnia disorder: a randomized controlled trial. J Affect Disord. 2026;122469. doi: 10.1016/j.jad.2026.122469
You must be logged in to post a comment.