PPubMed27 Mar 2019Cited 125×

Prediction of premature all-cause mortality: A prospective general population cohort study comparing machine-learning and standard epidemiological approaches Predicting premature all-cause mortality using machine-learning

Stephen F. Weng, Luis Vaz, Nadeem Qureshi, Joe Kai, Peyman Rezaei-Hachesu

Abstract

We found that machine-learning algorithms were better at predicting individuals who died prematurely, with higher discrimination, better calibration and classification accuracy, when compared to standard approaches. Deep learning based on a multi-layer feed-forward artificial neural network performed the best, with predictive accuracy improving by 3.9% compared to a multivariate Cox regression model, and by 10.1% using just age and gender alone. ML algorithms, when predicting vascular and non-vascular causes of death separately, demonstrated similar improvements in predictive accuracy. The stu

A figure from Prediction of premature all-cause mortality: A prospective general population cohort study comparing machine-learning and standard epidemiological approaches Predicting premature all-cause mortality using machine-learning — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

Machine‑learning models markedly improve prediction of premature all‑cause mortality over standard Cox regression in a large UK cohort.

What’s new

i.Developed novel machine‑learning prediction algorithms (random forest and deep learning) for premature all‑cause mortality
ii.Showed that ML models improve discrimination (AUC) by up to 10% compared with standard Cox regression
iii.Provided detailed reporting of ML integration with epidemiological study design, including hyper‑parameter grids and variable‑importance analysis
iv.Identified a reduced set of 15 key predictors that retain maximal discrimination

How they did it

A prospective cohort of 502,628 UK Biobank participants aged 40–69 (recruited 2006‑2010, followed to 2016) provided 60 baseline variables. The authors built Cox regression, random‑forest, and deep‑learning models, tuned via grid‑search and 10‑fold cross‑validation, then evaluated calibration and discrimination (AUC) on a held‑out test set of 125,657 individuals.

What they found

14,418 deaths (2.9%) occurred over 3,508,454 person‑years
Simple age‑gender Cox model AUC = 0.689; multivariate Cox AUC = 0.751 (6.2% improvement)
Random forest AUC = 0.783 (3.2% over multivariate Cox)
Deep learning AUC = 0.790 (3.9% over multivariate Cox)
ML models improved discrimination by 9.4% (RF) and 10.1% (DL) versus the simple Cox model
Deep learning sensitivity = 76.2%; random forest specificity = 77.5%

Caveats

✕ML algorithms act as “black‑boxes,” lacking clear effect‑size interpretation
✕Potential over‑fitting despite regularisation and hyper‑parameter tuning
✕Current ML models predict classification at follow‑up only, not time‑to‑event as Cox models do
✕Study uses internal data split and a “healthy volunteer” cohort, limiting external generalisability

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Prediction of premature all-cause mortality: A prospective general population cohort study comparing machine-learning and standard epidemiological approaches Predicting premature all-cause mortality using machine-learning

The paper, in five minutes

Ask a question about this paper

Similar papers in all four archives

The paper, in five minutes

Similar papers in all four archives