Thepprof software package provides a variety ofrisk-adjusted models for provider profiling, efficiently handlinglarge-scale provider data. It includes standardized measurecalculations, hypothesis testing, and visualization tools for evaluatingthe performance of healthcare providers and identifying significantdeviations from expected standards.
Provider profiling involves assessing and comparing the performanceof healthcare providers by evaluating specific metrics that reflectquality of care, efficiency, and patient outcomes. To achieve this, itis essential to fit statistical models and design appropriate measures.We developed thepprof package that facilitates fitting avariety of risk-adjusted models, each of which includes tools forcalculating standardized measures, conducting statistical inference, andvisualizing results, thereby offering a comprehensive tool for providerprofiling.
This package addresses key limitations in existing R functions forprovider profiling, which often suffer from computational inefficiencywhen applied to large-scale provider data. For the logistic fixed effectmodel, the serial blockwise inversion Newton (SerBIN) algorithm isimplemented, which leverages the block structure of the informationmatrix. For linear fixed effect models, a profile-based method is used.These, along with parallel computing capabilities, improve computationalspeed significantly.pprof handles diverse outcomes(e.g. binary and continuous) and offers both direct and indirectstandardization. pprof provides a comprehensive and user-friendly toolfor provider profiling, enabling users to fit risk-adjusted models,calculate standardized measures, perform hypothesis tests, and visualizeresults.
Note:The package is still in the early stagesof development, so please don’t hesitate to report any problems you mayexperience.
You can install ‘pprof’ via CRAN or github:
require("devtools")require("remotes")remotes::install_github("UM-KevinHe/pprof", ref = "main")If you encounter any problems or bugs, please contact us at:xhliuu@umich.edu,lfluo@umich.edu,kevinhe@umich.edu.
\[1\] Bates, D., Mächler, M.,Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects modelsusing lme4. Journal of Statistical Software, 67(1), 1-48.https://doi.org/10.18637/jss.v067.i01
\[2\] He, K., Kalbfleisch, J. D.,Li, Y., & Li, Y. (2013). Evaluating hospital readmission rates indialysis facilities; adjusting for hospital effects. Lifetime DataAnalysis, 19, 490-512.https://link.springer.com/article/10.1007/s10985-013-9264-6
\[3\] He, K. (2019). Indirect anddirect standardization for evaluating transplant centers. Journal ofHospital Administration, 8(1), 9-14.https://www.sciedupress.com/journal/index.php/jha/article/view/14304
\[4\] Hsiao, C. (2022). Analysis ofpanel data (No. 64). Cambridge University Press.
\[5\] Wu, W., Kuriakose, J. P.,Weng, W., Burney, R. E., & He, K. (2023). Test-specific funnel plotsfor healthcare provider profiling leveraging individual- andsummary-level information. Health Services and Outcomes ResearchMethodology, 23(1), 45-58.https://pubmed.ncbi.nlm.nih.gov/37621728/
\[6\] Wu, W., Yang, Y., Kang, J.,& He, K. (2022). Improving large‐scale estimation and inference forprofiling health care providers. Statistics in Medicine, 41(15),2840-2853.https://onlinelibrary.wiley.com/doi/full/10.1002/sim.938