Medical communication is bad–how can we fix it?
blog.tjcx.me
Medical communication is bad–how can we fix it?
1–3 of 3 posts
Re: Medical communication is bad–how can we fix it?
#2This is really well done. It's like a live-updating meta analysis. I wonder to what extent each paper can be parsed automatically to extract the p-value and effect size to add to the table.
I'm not sure what to do when there are methodological differences (e.g. randomized trials vs. observational studies), but I'm sure people who know more about meta-analysis have already developed techniques for merging such heterogeneous results.
Re: Medical communication is bad–how can we fix it?
#3This is neat. The author makes a list of problems with online medical information, defines what makes good medical information, and then coded a prototype to improve the accessibility of one class of medical information—the meta-analysis. The author's prototype imports 17 randomized, controlled trials on whether zinc lozenges reduce cold duration.
The prototype is interactive and you can add/remove trials to build your own informal meta-analysis. For example, I can easily generate screenshots that claim `According to 11 selected studies, there's an insignificant chance that zinc lozenges reduce cold durations by 1%.` or the exact opposite `According to 12 selected studies, there's a significant chance that zinc lozenges reduce cold durations by 21%.`. This provides the reader with better insight into the uncertainty of an issue, and also highlights the vulnerability of meta-analyses to manipulation. This tool, and some of the links provided by the author, highlight how higher level meta-analysis tools may build on the benefits offered by the original static meta-analysis.
The prototype also includes a "Funnel Plot", a clever chart type common in meta-analyses that I hadn't noticed much before.
One minor suggestion would be to include a 4th critical attribute of what constitutes "good" medical information: precision (aka personalization). A bit of demographic information can often make all the difference in a treatment. What has a 1,000% effect on one person might have predictably 0% affect on another, and so those decision trees should probably always be included. I suspect the author is aware of this and just cut it for scope.