I've been fortunate to review a number of papers on clinical prediction models this past year and I've noticed a number of frequent errors. This thread is mostly for the clinician-investigator considering building and publishing a predictive model of some kind. 3 major points. /1

@StephptaylorCLT Haha... I wish!

Very grateful to @pennaitech for the support, and to co-authors @LauraZwaan81 and Sigall Bell from whom I learned a ton in writing this!

RT @DrQiLong: Very excited to work with @RajeevAlur @AI4Code @RICEric22 @garyweissman @ravi_b_parikh @SameedKhatana Deo Rajat and Payal Sha…

RT @CatGaoHow: Great talk by @garyweissman at #AMIA2024 @AMIAinformatics - the time where broad spectrum abx should be started is not the s…

RT @PennMedCSO: Race & ethnicity & clinician linguistic expressions of doubt in hospital admission notes ft. @courtneley, Xinwei Chen (@Pen…

RT @PennLDI: For more on this, join us on October 10 for a day-long conference on GenAI in health care:

@callcura even if it had, we would have found very similar results with your initial PMN not including any peer reviewed manuscripts about the device or any rigorous assessments of clinical effectiveness, safety, or equity. Again, not specific to your company but with FDA PMN more broadly.

@callcura The summary statement doesn't mention any published studies about model performance let alone clinical effectiveness. The only cited studies in the summary statement describe the general problem of delayed imaging reads. Can you provide the citations for the studies you mention?

@callcura Are the methods & data for these performance estimates published? If so, those sources were not included in the FDA summary. Further, reporting predictive performance measures says nothing about clinical effectiveness. This is an issue with FDA regs not specific to your company.

@callcura Although if we had included this device ( that would have been available at the time of our data collection, it does not appear this PMN includes any published studies describing predictive performance, clinical effectiveness, safety, or bias assessments.

RT @_jcrojas: 🚨 Just released! 🚨 Excited to share our new preprint on the Common Longitudinal ICU Data Format (CLIF)! This open-source form…

RT @CALonghurst: Many thanks to the authors including @garyweissman of @PennLDI who provides further context and links to a github page for…

RT @CALonghurst: An editorial makes some specific recomendations including that the "FDA should require rigorous preapproval studies of val…

RT @CALonghurst: “Of 521 authorizations in the FDA AI/ML database, we identified 10 that might inform care for patients with critical illne…

RT @johnahartford: 📢 ANNOUNCEMENT | @theNAMedicine has selected 11 Scholars in Diagnostic Excellence in partnership with @CMSSmed for its 2…