New algorithm for constructing area-based index with geographical heterogeneities and variable selection: An...

Area of Science:

• Health economics
• Geospatial analysis
• Biostatistics

Background:

• Optimal health resource allocation requires accurate indicators of health inequality.
• Current area-based indices face challenges in incorporating geographical relationships and selecting variables from high-dimensional data.

Purpose of the Study:

• To develop a new area-based health coverage index using geographical information and variable selection.
• To characterize the geographical distribution of health inequality in Japan using gastric cancer as an example.

Main Methods:

• Proposed a geographically weighted logistic lasso model for index construction.
• Employed a geographical kernel and a two-stage algorithm to select optimal bandwidth and regularization parameters.
• Validated the index's sensitivity through correlation with cancer mortalities and screening rates.

Main Results:

• The developed index demonstrated a wider range (0.0001 to 0.354) compared to previous indices.
• Variable selection from 91 census data candidates showed regional differences, with a median of 29 variables selected.
• The new index exhibited stronger correlations with cancer mortalities/screening rates and identified geographical clusters with unique predictors.

Conclusions:

• The novel index effectively integrates geographical information and variable selection for a more accurate measure of health inequality.
• The findings reveal distinct geographical patterns of health inequality and unique local predictors in Japan.
• This approach offers a valuable tool for policymakers to optimize health resource allocation.