Predicting anaerobic threshold from warmup exercise by machine learning technique for daily appropriate exercise in patients with cardiovascular diseases
Y Suematsu, T Matsuda, S Takahashi, A Watanabe, K Fujimi, S I MiuraAbstract
Introduction
Anaerobic threshold (AT) which is an appropriate exercise intensity in patients with cardiovascular disease changes by physical condition, mental status, and the pathological condition. Excessive exercise intensity causes acceleration of sympathetic nerve activity, leads to angina, arrhythmia, and worsening of heart failure. Cardiopulmonary exercise test (CPX) with gas analyzer is a conventional method to detect AT, however, it is complicated and we have not measured AT in every exercise by this method.
Hypothesis
We developed a newly developed machine learning model to predict AT using a continuous blood pressure monitoring. A machine learning model based on long short-term memory with multi-head attention predicts the AT within discrete zones with a width of approximately 3 W (AT zone) using changes of the double product from rest to the warmup period (Figure). In this study, we investigated the differences between actual AT and predicted AT zone in healthy subjects and patients with cardiovascular diseases.
Methods
42 healthy subjects and 46 patients with cardiovascular disease were performed CPX with gas analyzer and non-invasive continuous blood pressure monitoring. CPX was performed with cycle-ergometer, followed by the protocol: a 4-min rest period, a 4-min warmup period, and an incremental exercise period with 10 W per minute increases. Actual AT was detected with conventional gas analysis by two independent evaluators. Predicted AT zone was derived from the developed model.
Results
In baseline characteristics, age, the percentage of male, and body mass index were 51.5±16.6 years, 67.0 %, and 22.5±3.1 kg/m2, respectively. In patients with cardiovascular disease, the percentages of ischemic heart disease, macrovascular disease, and heart failure were 58.7%, 6.5%, and 56.5%, respectively. The left ventricular ejection fraction was 51.5±16.5 % and the plasma level of brain natriuretic peptide was 118±130 pg/mL. In healthy subjects, the exercise intensity at actual AT point was 36.3±8.7 watt and the difference between actual AT and predicted AT zone was 6.3±6.9 watt. In patients with cardiovascular disease, the exercise intensity at actual AT point was 35.1±7.5 watt and the difference between actual AT and predicted AT zone was 4.3±5.2 watt. The differences between actual AT and predicted AT zone in both groups were acceptable for daily exercise prescription.
Conclusions
This exploratory research would be helpful for identifying AT for daily appropriate exercise without CPX in patients with cardiovascular disease, as well as in healthy subjects.Figure