How can doctors know which cancer treatment is better? One way is by riding slugs.

Learn how and pick the best cancer treatment option based on bleeding-edge scientific evidence from human trials: https://zheln.com/thread/2024/01/11/1/

If you feel critical about this post, we invite you to review its video bibliography, captured with @WriteInStonePublic, and drop any criticisms in the comments. Pavel Zhelnov, MD, will address all valid concerns and acknowledge the peer reviewers in the following Zheln post. We look forward to reading your messages.

Tamiflu is a widely advertised anti-flu drug that failed to show dramatic effects in clinical trials. Powered by @WriteInStonePublic, we revisit the evidence in light of a major 2023 meta-analysis and argue why highlighing failures might be as important as distilling benefits.

Seatbelt promotion works, but how well? 60% of drivers still don’t use seat belts, and even more passengers. Tracking steps with @WriteInStonePublic software, we discuss a recent systematic review that looked at the effectiveness of seat-belt promotion more closely.

Gentle human touch was shown to reduce pain in newborn babies. A new systematic review found that neonates do suffer from airway suctioning, but careful positioning likely alleviates pain. How did researchers measure it? Follow our steps using @WriteInStonePublic in a new Zheln post.

Zheln went #SocialFi, and here is a short post on opioid use disorder: https://hey.xyz/posts/0x04fcbd-0x05

What evidence do we have of the association between fluoride exposure and thyroid function? Check this funny summary of findings table: https://hey.xyz/posts/0x04fcbd-0x08-DA-a7d09302

Zheln’s (impolite) take on hazard vs. risk and some pieces of health research – and their authoritative academic publishers – that still fail to use a comprehensive search strategy in a systematic review and overlook the existence of ROBINS-E in 2024: https://hey.xyz/posts/0x04fcbd-0x09-DA-2c9ba1a1

Probiotics won’t make you poop, but lactulose will, with moderate certainty.

Taking lactulose for an average of 4 weeks leads to at least one more additional stool per week in adults with positive Rome criteria for constipation.

This post has been immortalized on a blockchain.

Can Universal Basic Income solve the problem of poverty? *

According to statistics, low income levels are associated with higher mortality rates and worse outcomes in the treatment of chronic diseases. Moreover, a report by the World Bank and World Health Organization shows that high healthcare costs push many people to the brink of poverty, forcing them to choose between buying food and paying for medical services.

To address this issue, the idea of introducing Universal Basic Income (UBI)—an untouchable sum granted to every member of society—has been discussed for many years. However, critics are skeptical of UBI, citing a lack of evidence on the strategic effectiveness of such a solution.

A new study published in the Campbell Systematic Reviews database sheds light on this issue. The authors conducted a meta-analysis to evaluate the impact of various forms of UBI on poverty-related indicators.

No skincare products proved effective against acne, a review finds. The market size of acne cosmetics is estimated at 4.5 billion U.S. dollars.

42,226 patients studied.

Zero media coverage.

Thoughts?

“Most companies do not reveal exact details of how their systems work.“ We do. This post announces a public test of Zheln AI for systematic reviews and explains how it works.

Zheln AI is a human-machine pipeline that curates published systematic reviews and other syntheses of health research to translate them into digestible news stories. Thus, it is not a tool to do a systematic review (sorry, fellow academics). Rather, it is a product for content creators who require a strong inflow of fair scientific evidence without sacrificing the process transparency.

The pipeline has three components: candidate report selection, report preprocessing, and summarization of evidence. The underlying software runs on premises, except for a few internet-dependent features, but can also be deployed to the cloud.

1. 👀 Candidate report selection

A comprehensive search for all systematic reviews is run in PubMed (and, in future, other sources). The search strategy has been public for years: https://doi.org/10.17605/OSF.IO/Z3JU7

The results are downloaded (about 10,000 a month), deduplicated with previous records, enriched with metadata from OpenAlex, and filtered. We’ve not made the algorithm public yet, but a lot is shown in the @WriteInStonePublic reports on Zheln.com

Shortlisted records are displayed for human review. To assist with selecting the reports of interest, there is an AI query box that can sort the candidate list based on a free-text prompt (for geeks: rerankers answerdotai-colbert-small-v1).

There will also be an AI peer review module, which we are currently testing, that can assess the relevance of the study from a certain perspective, e.g., that of a patient advocate or a head of surgery in a particular disease field. Follow researchwaste.info for updates on this.

2. ✍️ Report preprocessing

In a separate tab, the user can collect and collate all the pertinent reports related to the selected studies. The user can review and record the license under which the full text is distributed before any text mining takes place. About 30% of all systematic reviews published today go under a license that allows text mining for commercial purposes (e.g., CC-BY), and another 20% permit non-commercial AI use.

To assist with the report collection, Zheln AI includes a browser tool that can grab the full text from the publisher’s web page, and there is also a module that can automatically collect the PDFs that are freely available on the internet. The full text is saved to the tool’s database.

3. 🤖 Summarization of evidence

Here is the juicy part. Over the years, we at Zheln have summarized hundreds of systematic reviews. Now we leverage this dataset as context for a large language model to have AI do the job (the choice of LLM provider depends). The full text report is used whenever available; otherwise, the abstract is summarized.

—
⭐️ Sneak peek! Leave a comment with a paper or topic you’d like to run through Zheln AI, and we’ll post a screen-capture in reply. 👇

🔎 Here is a sneak peek of what @tgzheln new AI pree review module can do (spelling intended). 👩‍💼👨🏽‍⚕️🧑‍🎓

🖼️ The attached screenshot is a critical summary of a recent systematic review by Sana Mohammadi et al., performed with a large language model from a patient advocate’s perspective. A text version is in the comments!

👇 You may leave a topic or article you would like to run through the tool in the comments below, and we’ll post a @WriteInStonePublic desktop capture in response.

🔄 Shares are appreciated!