Integrated artificial intelligence in healthcare and the patient's experience of care

Abstract

Healthcare is plagued with many problems that Artificial Intelligence (AI) can ameliorate or sometimes amplify. Regardless, AI is changing the way we reason towards solutions, especially at the frontier of public health applications where autonomous and co-pilot AI integrated systems are now rapidly adopted for mainstream use in both clinical and non-clinical settings. In this regard, we present empirical analysis of thematic concerns that affect patients within AI integrated healthcare systems and how the experience of care may be influenced by the degree of AI integration. Furthermore, we present a fairly rigorous mathematical model and adopt prevailing techniques in Machine Learning (ML) to develop models that utilize a patient's general information and responses to a survey to predict the degree of AI integration that will maximize their experience of care. We model the patient's experience of care as a continuous random variable on the open interval ([Formula: see text]) and refer to it as the AI Affinity Score which encapsulates the degree of AI integration that the patient prefers within a chosen healthcare system. We present descriptive statistics of the distribution of the survey responses over key demographic variables viz. Age, Gender, Level of Education as well as a summary of perceived attitudes towards AI integrated healthcare in these categories. We further present the results of statistical tests conducted to determine if the variance across distributions of AI Affinity Scores over the identified groups are statistically significant and further assess the behavior of any independent distribution of AI Affinity Scores using a Bayesian nonparametric model.

Keywords: AI and patient care; AI in public health; Deep learning in healthcare.

Fig. 1

Normalized KDE plot for affinity scores across demographic groups.

Fig. 2

Box plot for affinity scores across demographic groups.

Fig. 3

Digital approval level across age and region groups.

Fig. 4

AI support level across age and region groups.

Fig. 5

Training error surface.

Fig. 6

Confusion matrix.

Fig. 7

Distribution of predicted versus observed AI affinity scores from the deep learning model and the linear regression baseline model (N = 320).

Fig. 8

Predicted versus observed affinity scores.