Lee Y; Department of Radiological Sciences Medical & Imaging Informatics (MII) Group Los Angeles CA
Feng J; Rahrooh A; Bui AAT; Cooper CB; Hsu JJ

Journal of the American Heart Association. 15(6):e045734, 2026 Mar 17.

BACKGROUND: Cardiorespiratory fitness, as measured by peak oxygen uptake
during cardiopulmonary exercise testing, is a prognostic indicator. We aim
to predict peak oxygen uptake from submaximal variables on cardiopulmonary
exercise testing to assess cardiorespiratory fitness when maximal exertion
is not possible.

METHODS: Data from 13535 cardiopulmonary exercise testings were
collected, and patients were divided into a normal group (NG; n=1076) and
other group (OG; n=9823). Regression models to predict maximum oxygen
consumption were trained and evaluated on the NG, OG, and combined groups
(NG+OG) using stratified 5-fold cross-validation. We trained different
models using demographic, resting and submaximal variables.

RESULTS: Optimal models were Bayesian Ridge for the NG and Light Gradient
Boosting Machine for the other groups. The mean (SD) R2 when using
demographic and rest variables was 0.690 (0.027) for the NG, 0.546 (0.012)
for the OG, and 0.562 (0.015) for the NG+OG. When using demographic, rest
and submaximal variables, performance increased to 0.796 (0.020) for the
NG, 0.732 (0.009) for the OG, and 0.761 (0.008) for the NG+OG. Oxygen
consumption at the first ventilatory threshold, minute ventilation at the
second ventilatory threshold, and forced expiratory volume in 1 second
were important features across the models trained with rest and submaximal
variables. Minute ventilation at the second ventilatory threshold had
negative effects, while oxygen consumption at the first ventilatory
threshold and forced expiratory volume in 1 second had positive effects on
maximum oxygen consumption prediction. In exploratory analyses, the
inclusion of chronotropic index improved model performance.

CONCLUSIONS: Our peak oxygen uptake prediction model demonstrated strong
performance using submaximal exercise variables. This methodology offers a
means to assess prognostic markers for individuals who might not achieve
maximal exhaustion during cardiopulmonary exercise testing.