Thread by @MaartenvSmeden

This my *top 10* of favorite methods papers of 2019

Appearing in a single thread and in random order

Disclaimer: this top 10 is just personal opinion; I’m biased towards explanatory methods and statistics articles relevant to health research

#1: A plea against dichotomization of study results as statistically significant or not. With three authors, 800+ signatories and #2 Altmetric score of the year, this article belongs on the list. Agree with the message or not, or only a little

go.nature.com/2Qd0eUI

#2: “Machine learning” does not always outperform the predictive abilities of default statistical models (or vice versa), especially when data are small and noisy. This much discussed systematic review of clinical prediction models shows exactly that

bit.ly/2MmPBNO

#3: Deep learning to replace medical doctors in detecting disease from medical images? This systematic review identified lots of publications and promises but, so far, disheartening low levels of evidence

bit.ly/2rjWSGW

#4: Something more positive about machine learning? This 5-article series in Biostatistics about machine learning for causal inference written by early- and mid-career statisticians

bit.ly/36TNwRs

#5: Cancer or not cancer? With some lesions that are in the “gray zone” that is sometimes difficult to judge. Algorithms suffer from the absence of a gold standard as this recent article in @NEJM eloquently argues

bit.ly/2Qc13Nt

@NEJM #6: Multiple imputation has become *the* standard tool for fixing missing data problems in health research. But should it always be used when, say, more than 10% of the data are missing? Possibly not

bit.ly/34PjywD
bit.ly/2EHPsAx

@NEJM #7: The widely used measure of potential impact of treatments and interventions, the number needed to treat (NNT), should probably be treated with a healthy bit a skepticism. This article lays out the arguments

bit.ly/2QpvO1J

@NEJM #8: Sample size for prediction models have long relied on rules-of-thumb. No longer. These two articles come with new approaches and software (Stata and R) for calculating the minimal sample size

bit.ly/2sTzVuM
bit.ly/2SftRaB

@NEJM #9: How to select confounders to control for in applied health research? This article lays out the state of the art in confounder selection

bit.ly/2PPZMN4

@NEJM #10: Finally, the publication of an article pleading for post-hoc power calculations in a surgical journal got #statstwitter very excited in 2019. Leading to numerous twitter threads, blogs, 45 replies on @PubPeer and some letters to the editor

bit.ly/2Ziws50