--- title: "Best Practice Research" author: "Jenna Reps, Peter R. Rijnbeek" date: '`r Sys.Date()`' output: rmarkdown::html_vignette --- ```{=html} <!-- %\VignetteEngine{knitr::rmarkdown} %\VignetteIndexEntry{Best Practices} %\VignetteEncdoing{UTF-8} --> ``` ## Best practice publications using the OHDSI PatientLevelPrediction framework <table> <tr> <th> Topic </th> <th> Research Summary </th> <th> Link </th> </tr> <tr> <td> Problem Specification </td> <td> When is prediction suitable in observational data? </td> <td> Guidelines needed </td> </tr> <tr> <td> Data Creation </td> <td> Comparison of cohort vs case-control design </td> <td> <a href='https://doi.org/10.1186/s40537-021-00501-2'>Journal of Big Data</a> </td> </tr> <tr> <td> Data Creation </td> <td> Addressing loss to follow-up (right censoring) </td> <td> <a href='https://doi.org/10.1186/s12911-021-01408-x'>BMC medical informatics and decision makingk</a> </td> </tr> <tr> <td> Data Creation </td> <td> Investigating how to address left censoring in features construction </td> <td> <a href='https://doi.org/10.1186/s12874-021-01370-2'>BMC Medical Research Methodology</a> </td> </tr> <tr> <td> Data Creation </td> <td> Impact of over/under-sampling </td> <td> <a href='https://doi.org/10.1186/s40537-023-00857-7'> Journal of big data</a> </td> </tr> <tr> <td> Data Creation </td> <td> Impact of phenotypes </td> <td> Study Done - Paper submitted </td> </tr> <tr> <td> Model development </td> <td> How much data do we need for prediction - Learning curves at scale </td> <td> <a href='https://doi.org/10.1016/j.ijmedinf.2022.104762'>International Journal of Medical Informatics </a> </td> </tr> <tr> <td> Model development </td> <td> What impact does test/train/validation design have on model performance </td> <td> <a href='https://doi.org/10.1136/bmjopen-2021-050146'>BMJ Open </a> </td> </tr> <tr> <td> Model development </td> <td> What is the impact of the classifier </td> <td> <a href='https://doi.org/10.1093/jamia/ocy032'>JAMIA</a> </td> </tr> <tr> <td> Model development </td> <td> Can we find hyper-parameter combinations per classifier that consistently lead to good performing models when using claims/EHR data? </td> <td> Study needs to be done </td> </tr> <tr> <td> Model development </td> <td> Can we use ensembles to combine different algorithm models within a database to improve models transportability? </td> <td> <a href='https://doi.org/10.3233/SHTI230080'> Caring is Sharing–Exploiting the Value in Data for Health and Innovation </a> </td> </tr> <tr> <td> Model development </td> <td> Can we use ensembles to combine models developed using different databases to improve models transportability? </td> <td> <a href='https://doi.org/10.1186/s12911-022-01879-6'> BMC Medical Informatics and Decision Making </a> </td> </tr> <tr> <td> Model development </td> <td> Impact of regularization method </td> <td> <a href='https://doi.org/10.1093/jamia/ocae109'> JAMIA </a> </td> </tr> <tr> <td> Evaluation </td> <td> Why prediction is not suitable for risk factor identification </td> <td> <a href='https://proceedings.mlr.press/v182/markus22a.html'> Machine Learning for Healthcare Conference </a> </td> </tr> <tr> <td> Evaluation </td> <td> Iterative pairwise external validation to put validation into context </td> <td> <a href='https://doi.org/10.1007/s40264-022-01161-8'> Drug Safety </a> </td> </tr> <tr> <td> Evaluation </td> <td> A novel method to estimate external validation using aggregate statistics </td> <td> Study under review </td> </tr> <tr> <td> Evaluation </td> <td> How should we present model performance? (e.g., new visualizations) </td> <td> <a href='https://doi.org/10.1093/jamiaopen/ooab017'>JAMIA Open</a> </td> </tr> <tr> <td> Evaluation </td> <td> How to interpret external validation performance (can we figure out why the performance drops or stays consistent)? </td> <td> Study needs to be done </td> </tr> <tr> <td> Evaluation </td> <td> Recalibration methods </td> <td> Study needs to be done </td> </tr> <tr> <td> Evaluation </td> <td> Is there a way to automatically simplify models? </td> <td> <a href='https://ohdsi-studies.github.io/FeatureSelectionComparison/docs/Protocol.html'>Study protocol under development </a> </td> </tr> </table>