A Preliminary Report on Predicting Asthma Severity Level and Asthma Attack - A Deep Learning Approach

Quan T Do, Son C Tran, and Jill A. McDonald


smart health, big data analysis, deep learning, neural networks, wireless health


Assessing and monitoring asthma severity is the key for asthma treatment. To classify patients’ asthma severity level, the dataset of all asthma patients, which is extracted from the United States (US) largest publicly available all-payer inpatient health care database (NIS 2012) is used to train the machine using a Deep Learning (DL) algorithm. In the study being reported nine strong predictors include age, admission month, number of procedures, gender, race, number of chronic conditions, drug(s) in use on discharge date, length of stay in hospital, and mean income in the zip code of the patients’ location, were delineated. The DL classification accuracy rate using these 9 predictors was, nearly, 69%. In order to improve the accuracy rates of classifying asthma severity level and prediction of near future asthma attack, other data such as the weather-health information (e.g. humidity, pollen, air quality), asthma self-management data (e.g. symptoms, medications, triggers, comorbidity), the sensory Peak Expiratory Flow and pulse-oximetry will be added to the list of input variables. This research is still in progress; the self-learning model to increase the prediction accuracy rate by learning from data accumulated over time will be investigated, following by a randomized clinical study.

Important Links:

Go Back