Predicting Fluctuating Rates of Hospitalizations in Relation to Influenza Epidemics and Meteorological Factors

Area of Science:

• Epidemiology
• Public Health
• Environmental Health

Background:

• Influenza epidemics significantly increase hospital admissions in France.
• Accurate prediction of influenza-associated hospitalizations is crucial for anticipating healthcare demand.
• This study investigates predictors of influenza epidemics using meteorological and medical data.

Purpose of the Study:

• To identify key predictors of influenza epidemics.
• To analyze the relationship between meteorological factors, general practitioner data, and hospitalizations.
• To develop a predictive model for influenza-associated hospital admissions.

Main Methods:

• Collected 8 years of historical data (2007-2015) on meteorological factors, general practitioner consultations (Sentinelles network), and hospital admissions.
• Utilized Pearson correlation, Principal Component Analysis, and k-means clustering to assess data relationships.
• Employed linear discriminant analysis for epidemic state prediction and Markov chains for transition probability calculations.

Main Results:

• High correlations were observed between influenza hospitalizations and general practitioner activity (Sentinelles network, emergency admissions).
• Significant anti-correlations were found between hospitalizations and meteorological factors (temperature, absolute humidity, solar radiation) with specific time lags.
• Linear discriminant analysis accurately predicted epidemic weeks, with transition probabilities to epidemic states strongly linked to thresholds of meteorological variables.

Conclusions:

• Confirms a strong correlation between influenza-associated hospitalizations, meteorological conditions, and general practitioner activity.
• General practitioner activity emerged as the most significant predictor of hospital activity during influenza epidemics.
• Identified specific meteorological thresholds predictive of increased influenza epidemic risk.