Devin Incerti

@joeymattingly @brootmcq @mikepaulden @djvanness @DARTHworkgroup @AnkurFactorial @DrWmPadula @juliaslejko We're moving code over to Python so hopefully it won't be considered too wild in the near future 😁

@brootmcq @mikepaulden @djvanness @DevinIncerti @DARTHworkgroup @AnkurFactorial @DrWmPadula @juliaslejko What programs are you covering for modeling? You going with something old school like Excel? User-friendly like TreeAge? Open-source like R? Or maybe something wild like Python?!? 🤓

External controls are intriguing but bias is a concern. Across 14 old studies, we found that trial controls survived longer---even after PS adjustment---than external controls (HR=0.90). We propose a meta-analytic framework to adjust for such bias. arxiv.org/pdf/2110.03827…

Big thanks to the glmnet team for making it possible to model start/stop survival data and adjust for left-truncation. #cox-models-for-start-stop-data-1" target="_blank" rel="nofollow noopener">glmnet.stanford.edu/articles/Coxne…

We think this type of data will be increasingly common as datasets that link genomic information with clinical outcomes become more common. All code for our analysis is available in our GitHub repo.github.com/phcanalytics/c…

Happy to see our paper w/@EsMagoo,@robtibshirani & all on prediction modeling with high-dimensional left-truncated and right-censored survival data published in Statistics in Medicine. Need to be especially careful using C-index for model evaluation. onlinelibrary.wiley.com/doi/full/10.10…

@JonathanBriody @JeroenPJansen @ISPORorg Thanks so much @JonathanBriody, I'm glad you found the materials helpful! And hope you will consider R for your cost effectiveness modeling 😃

Thanks to @DevinIncerti, @JeroenPJansen and of course @ISPORorg for the short course on Applied Cost-Effectiveness Modelling with R. The R code on probabilistic sensitivity analysis was particularly appreciated (with apologies to team Stata!) #EconTwitter #AcademicTwitter

New version of hesim (0.5.1) on CRAN. Updates include new summary methods for checking/summarizing model inputs, enhancements to make it easier to build large transition matrices, and improved documentation, among others. #hesim-051" target="_blank" rel="nofollow noopener">hesim-dev.github.io/hesim/news/ind…

Looking forward to teaching an @ISPORorg short course (Aug 11/12) on cost-effectiveness modeling with R with @JeroenPJansen. We'll build models in both base R and hesim and have made our course materials available online. ispor.org/conferences-ed… github.com/hesim-dev/rcea

@ajhatswell @waq0r Agree that the main advantage of R is the stats integration

Why is no one building health economic models in JavaScript? It's 100x faster than R (presumably 10,000x faster than MS Excel), much easier to write than C, and seamlessly integrates with interactive web applications.

@waq0r @ajhatswell Certainly if you want someone to be able to run something interactively on the web than a few hours isn't going to cut it. It can also be a pain to debug if run time is on the order of hours. Likewise sensitivity analyses get annoying with slow run times.

@ajhatswell Right, when the gain is in the magnitude of minutes or hours, it's probably not worth it. Even when it's days, you could just throw more hardware at the problem. The RCA of JS might then only be in web interface integration +/- really complicated models.

@felbalazard @OWKINscience @Inria @Univ_Paris Much appreciated @felbalazard. Thank your for putting together such a great event!

Great talk by @DevinIncerti on meta-analytical approaches to external control arms at leveraging observational data workshop @OWKINscience @Inria @Univ_Paris . RWD patients tend to die sooner and this can be estimated by using reference studies

@rhta16 @feralaes @leatl4 Thanks @rhta16, really excited to be joining!

And speaking of new members of the SciCom, welcome to @feralaes, Raquel Aguiar-Ibanez, @DevinIncerti, Dawn Lee and @leatl4!

@djvanness There is a pretty simple example in the Gelman and Hill textbook that uses simulation. I think it would extend in a straightforward way to survey weights. The alternative is to use the delta method to get standard errors, but I think simulation is easier.

@brootmcq Could also be the quality/clarity of the code that was written. Definitely worth learning Python though!

It's still slower than simulating from a parametric model because we use the inverse CDF method for random number generation and the quantile function must be computed numerically, but the performance penalty is relatively small.

Some new benchmarks for microsimulations parameterized with spline-based survival models in our #rstats hesim package. #Rcpp makes it surprisingly fast: in realistic settings, run time is half a minute compared to 3.5 hours with other packages! hesim-dev.github.io/hesim/dev/arti…

@brootmcq @JeroenPJansen Thanks @brootmcq!

New preprint with @JeroenPJansen describing our hesim R package for fast simulation of cost-effectiveness models. Accompanies release of v0.5.0 on CRAN with many new features. arxiv.org/abs/2102.09437

Uncovering immortal time bias when fitting prognostic survival models in a large clinico-genomic database. Here, our assessment w/ @DevinIncerti, @robtibshirani & all (pre-print). medrxiv.org/content/10.110…