Continuing on with added context: The reason this one paper matters so much is its recency. EviCore has, for many years now, deemed LDRT for OA as medically necessary. To reverse course means new information must have become available between their last version and this one. This Hammadeh paper was published in September 2025, which means it technically predates the guidelines from February 2026 covering LDRT, but...we'll just say EviCore didn't find it in time for those guidelines. Unfortunately, it's not "new" data to justify this reversal, it's just a meta-analysis. But it has a convenient conclusion (if you're an insurance company). Alright, we can skip over the publishing Round Robin AI Ring the authors have going and just head right to the first graphic. Table 1. So, Table 1, second column. Several of the study design labels are incorrect - RCTs listed as cohort, retrospectives listed as RCTs, etc. Ultimately, not SUPER problematic. But uh, the next column is. The number of patients for each study in the third column. Those are not anywhere close to the patient populations actually used in those studies. The difference is severe (Hermann 2021 is listed as 160, actual is 25; Keller 2013 is actually over a thousand, listed as 120, etc). Well that's not good. And then we get to "Duration of Complaints". Somehow, mysteriously, all of these studies enrolled patients with 9.5-11.1 years of symptoms prior to treatment. Obviously...wildly incorrect as well. But the best are the final two columns which are outcomes. Second to last is pain, last is function. It sure is weird that, for a meta-analysis claiming LDRT has no effect, that each and every study allegedly used in their analysis ALL show an improvement in pain and functional status. Now, normally I would expect a meta-analysis to conclude positive treatment effect if, you know, each and every included study showed a benefit in multiple domains. Obviously...those results aren't what the original manuscripts themselves concluded (because, for example, they might have measured outcomes completely differently...). Oh and I totally forgot the column that somehow claims each included study miraculously enrolled 38-44% males. All of them. Crazy! The one outlier is the final study, Fazilat-Panah et al, 2025. If you look at the history of this Hammadeh paper, it was submitted originally in March 2025 and accepted in July 2025. The Fazilat-Panah paper was published online in March 2025. Obviously, Hammadeh and friends couldn't have included the Fazilat-Panah paper with their original submission - they wouldn't have known about it. So one of the reviewers (hard to imagine this paper had reviewers, but whatever) must have asked them to include this newly published paper as a revision. But that Fazilat-Panah paper is excellent: it's a double-blind sham-controlled RCT showing a tremendous benefit of LDRT. Thus, when Hammadeh plugged that paper in and re-ran his AI tool...that's what caused the the calculated I² to be a massive 96%. My assumption is the AI tool hallucinated much of the original data towards "no benefit". But when this Fazilat-Panah paper data was included after the fact with the real results - the heterogeneity went off the rails. No paper should be published with an I² of 96%. So, I assume no reviewer actually read this paper after it was re-submitted as a revision. Then the editors rubber-stamped the AI trash, and boom! @evicorehc and their AI scraping algorithm programmed to find any and every reason for denial uncovered this gem and here we are today. Who doesn't love science and medicine in 2026? (probably the suffering patients getting treatments denied don't love it, but EviCore hasn't ever cared about that before, and neither has @SpringerNature)