Detecting the patient's need for help with machine learning based on expressions

Area of Science:

• Health analytics
• Machine learning
• Statistical analysis

Background:

• Machine learning models for health analytics require understanding statistical properties of self-rated health statements.
• Research analyzes COVID-19 related self-rated statements to link group differences to machine learning outcomes.

Purpose of the Study:

• To analyze self-rated expression statements about the COVID-19 epidemic.
• To develop and apply a new methodology for influence analysis in machine learning.
• To link statistically significant differences in respondent groups to machine learning results.

Main Methods:

• Quantitative cross-sectional study with 673 online respondents from Finnish organizations.
• Collected "need for help" ratings on 20 health statements (11-point Likert scale) and background data (health, wellbeing, sex, age).
• Proposed and tested a new machine learning influence analysis methodology.

Main Results:

• Found significant Kendall rank-correlations and high cosine similarity between statement ratings and background questions.
• Identified significant rating differences based on health condition, quality of life, and sex using Wilcoxon, Kruskal-Wallis, and ANOVA tests.
• Demonstrated how statistically significant rating differences link to machine learning results.

Conclusions:

• Self-rated "need for help" varies significantly with individual background information (health, quality of life, sex).
• The new methodology connects significant rating differences to machine learning outcomes.
• Enables development of improved machine learning for personalized patient care.