Lab-Life

In a recent paper, Tinglong Dai, Risa Wolf, and Haiyang Yang wrote about unlearning in Medical AI. With more and more CPU and storage thrown at Large Language Models (LLMs) and Generative Artificial Intelligence (GenAI) in general, the capacity to “memorise” information grows larger and larger with each generation of LLM. This is further accelerated by the capacity to add specific details to generalist LLMs using Retrieval Augmented Generation (RAG) and agents (e.g., with the ability to query real-world systems at the interface with the physical world). LLMs are learning more, but what about unlearning? Dai and colleagues didn’t evoke the analogy with human memory: our capacity to learn more relies, in part, on our capacity to forget, to reorganise, to summarise, and to prioritise the learnt information. Sleep and stress play a role in this reorganisation of information; this was the overarching topic of my Ph.D. thesis [link]. I will de-prioritise the visual cues along the path leading to a bakery if I no longer go to this bakery (“unlearning”). However, practising navigation to the bakery improved this skill, and this improvement will serve me later when I need to go to another place (something I could call “secondary learning”). It may seem we diverge from AI, but Dai and colleagues actually start their paper with the EU GDPR possibility for a patient to remove their data from a database, wondering how this is technically possible with LLMs (where data is not structured like in a traditional relational database and where the way data is retrieved is often unknown). The “unlearning” process in LLMs can be considered from three encapsulated levels: algorithm, legal, and ethical levels.

After Modelling and Regulations & Pricing, and just a few days before ISPOR25, here is my take on the potentially interesting sessions on Artificial Intelligence (AI, which generally means: the use of Generative AI, or GenAI, in HEOR). First, Sven Klijn, William Rawlinson, and Tim Reason are again offering their introductory course on Applied Generative AI for HEOR. Last year, I followed it in Barcelona, and it was nice. In my opinion, “nice” means that although I didn’t learn much more than previous presentations by the authors and my own experience, it was a great course for beginners because it struck the right balance between theory (which too many sessions end up only covering) and practical examples. Don’t expect hands-on exercises (that would be too long, and the course synopsis doesn’t mention that either). But “nice” to me means that the presenters dared to show actual working code, with all the humility that it implies. This year, they mention they’ll cover Retrieval-Augmented Generation (RAG) and agents. Hopefully, their coverage of these aspects will be as good as last year’s on the other topics. ...

ISPOR25, the annual North American conference for the International Society for Pharmacoeconomics and Outcomes Research, is in three weeks. As usual, I’m planning for it by browsing its program. This time, I decided to share a few of my interests on my blog. ISPOR usually covers many topics, from “hardcore” statistical methods to top-level overviews of some issues, so I will focus on only a few topics. Feel free to connect with me if you want to discuss anything at or around the conference (or virtually). (And before we start, full disclaimer: I’m currently working for Parexel, but opinions shared here are only mine; otherwise, I would have written them on the company blog.) ...

Five months ago, I was wondering why Maryland remove COVID-19 cases from its count in congregate living facilities (nursing homes, prisons, …). I still don’t have any answer but I found a technical solution :-) The Python script (in src/ in the MD-coronavirus repo on Github) just fills in the latest data for days where data is missing. On a side note, it also fix some basic issues like a reporting date in year “0200” (instead of “2020”). You can play with the fixed data file here. ...

Recently (I’m writing this on October 20), the (new) Belgian government decided to apply more stringent prophylaxis measures to contain COVID-19. One of the controversial measure is to close bars and restaurants for a month. Unfortunately, in a way, at approximately the same time, AVIQ released its latest poll on COVID-19 clusters in Wallonia ( AVIQ is the Walloon agency for well-being, health, handicap and family). I wrote it was unfortunate because I read and heard several people who criticized the closing of bars and restaurants by citing this poll. But this poll cannot answer in favor or against this closure; it doesn’t look at that … ...

Sometimes, you think that you found something interesting but the Maryland Department of Health is already presenting it on its COVID-19 dashboard :-D For instance, I calculated the percentage of residents of the different counties ever tested (regardless of the test result). I found out that a third of Maryland counties (8/24) tested at least once more than 25% of their residents. Indeed, as of yesterday (August 10), here are the counties in that category: ...

Since mid-July 2020 in Maryland, we understood that the 20-59 yr age group was problematic, especially the 20-29 yr age group that is racing to overtake all age groups in terms of number of COVID-19 cases (relative to their population, see top chart below). In terms of COVID-19 hospitalizations, we also saw a small rebound (see chart below; it seems that it subsides since beginning of August). But what we didn’t know (for this small peak as well as since the beginning) was what is the age of these hospitalized populations. Did these hospitalizations impacted more the older adults? The younger ones? Or the children? The Maryland Department of Health COVID-19 dashboard doesn’t report that information (nor in the API). ...

Since the beginning of this pandemic, one metric intrigued many of us in Maryland: the cumulative number of people released from isolation. Initially (before the data release via API, when there was only the MDH dashboard), it was even thought to be the number of hospital patients released from isolation. It’s not: the API page mentions: Total Number Released from Isolation data layer is a collection of the statewide cumulative total of individuals who tested positive for COVID-19 that have been reported each day by each local health department via the ESSENCE system as having been released from home isolation. As “recovery” can mean different things as people experience COVID-19 disease to varying degrees of severity, MDH reports on individuals released from isolation. “Released from isolation” refers to those who have met criteria and are well enough to be released from home isolation. Some of these individuals may have been hospitalized at some point. ...

I may have missed when the data was first released but I discovered the Maryland Department of Health (MDH) is publishing some data about COVID-19 contact tracing (in Maryland). This data is not on the main COVID-19 dashboard but on the contact tracing page (and in the datasets that can be downloaded). Here is a first insight of what happened so far … Note: if you just look for where to get tested in Maryland, the official information is here. ...

Every day, Governor Hogan and members of his team are communicating news on the COVID-19 situation in Maryland via Twitter (and other media): @GovLarryHogan, @riccimike, @katadhall, … (and of course: @MDHealthDept too!). A number of data enthusiasts are also parsing the MD Department of Health data: @TylerFogarty7, @MikeBReporter, @JauquetW, @PrayagGordy, … and of course: @jepoirrier) … And this is only on Twitter! But also every day, there is one thing that constantly changes: how everyone is calculating the COVID-19 positivity rate. Today (July 26), for instance, the different daily positivity rates announced are: 3.77% ( Hogan), 4.47% ( Hogan again in the same tweet, Hall, Ricci, MD Health Department, Fogarty) and ~6% (for me, the exact number behind the ~ is 6.14%). This doesn’t show the 7-days (or n-days) averages and other measures. And this is only on Twitter. ...