Evaluating a logistic regression based prediction tool in R

This is a tutorial on how to use R to evaluate a previously published prediction tool in a new dataset. Most of the good ideas came from Maarten van Smeden, and any mistakes are surely mine. This post is not intended to explain they why one might do what follows, but rather how to do it in R. It is based on a recent analysis we published (in press) that validated the HOMR model to predict all-cause mortality within one-year of a hospitalization in a cohort of patients aged 65 years or older that were under the care of geriatric medicine service at Cork University Hospital (2013-01-01 to 2015-03-06).

Science doesn't have a problem

Everywhere I look, people are saying there is something wrong with Science.

Dealing with negative trial results

Welcome to the world of clinical trials. You’ve got a great idea that has real potential to improve health. But I have some bad news for you. Are you ready?

No, we can't censure people for ad hominem attacks in scientific discourse.

“Individuals engaging in ad hominem attacks in scientific discourse should be subject to censure.” From Issues with data and analyses: Errors, underlying themes, and potential solutions. Andrew W. Brown, Kathryn A. Kaiser, and David B. Allison. PNAS March 13, 2018. 115 (11) 2563-2570; published ahead of print March 12, 2018. That is a remarkable suggestion.

Magical clusters

I was greeted today with the news that there are 5, not 2, types of diabetes. This is earth-shattering if you are a diabetologist or diabetes researcher. However, I soon as I saw the term “data-driven clustering” I knew I could probably relax. For the uninitiated, data-driven clustering techniques can seem magical. The basic premise is that you take a sample, measure lots of things about them, and then feed all those data into an algorithm that will divide your sample into mutually exclusive groups.

Can social media be useful for scientists?

I have been blogging and using Twitter as a scientist since 2010. By that point it was pretty obvious that internet was radically changing how scientists could engage with the public and each other, and thus science blogging had become quite popular. Like a lot of people, I wanted to write about how studies from my areas of expertise were reported in the media, and my first substantial blog post was about a large epidemiological study of homebirths in the UK. Twitter was a natural companion to blogging, since you could use it to share what you were writing. Eight years later I still blog and tweet. From time to time this comes up in conversation with colleagues, and they often ask if I really think it’s worth my time.

New workshop for citizen scientists

I am happy to announce that in the new year we will be running a series of workshops for Cork Citizen Scientists. In contrast to the small series of lectures we ran this year, the new workshops will be completely focused on supporting local citizen scientists in their efforts to answer real questions with real data.

Sex differences

Remember the whole Google Memo thing that happened a hundred years ago? Its central argument (as far as I could tell) was that the large male to female sex ratios we observe in Tech can be reasonably explained by small differences in the sex-specific probability distributions of innate characteristics. Thus Google’s attempts to increase diversity were silly at best, perhaps harmful or unjust, and largely due to a culture of political correctness that was oppressive to “conservative viewpoints”.

Response Heterogeneity

A colleague in food science recently sent me a narrative review outlining some of the challenges in their field. One of these was “extensive heterogeneity in the response to increased intake [of flavonoids]”. So-called response heterogeneity is often highlighted to justify the need for precision medicine, but there is a problem with this: the studies that are used to demonstrate response heterogeneity simply don’t.

Rater effects

I was recently asked to help analyze some assessment data. There were 50 people applying for seven positions. Each person’s application materials were scored by three people, randomly chosen from a larger group of raters. I was asked to help account for the fact that some raters might have a tendency to give higher or lower than average scores. If you want to play along, you can download the data here: