Testing and evaluating updated variants of the rurality index for Japan: a nationwide methodological study

To update the rurality index for Japan (RIJ) using the most recent national data and to test and evaluate the updated RIJ variants that incorporate alternative distance metrics and a modified classification of remote islands, thereby providing methodological guidance for their use in research and health policy.

Nationwide methodological study.

Japan.

All postal code areas in Japan; analyses were aggregated into municipalities and secondary medical care areas for evaluation.

Six RIJ variants ('RIJ family') constructed by combining three distance metrics (direct distance, road-based distance and travel time by car) and two island classifications (original RIJ and modified RIJ), standardised to a continuous 0-100 scale using percentile ranks.

Concordance among RIJ variants was assessed using Spearman's rank correlation coefficients, quintile reclassification matrices and rank-difference analyses; convergent validity was evaluated using the national physician distribution index. Criterion-related validity was assessed using municipal-level life expectancy for men and women.

All RIJ variants had extremely high concordance (Spearman's ρ≥0.99 across all pairwise comparisons). Compared with the previous RIJ classification, >98% of the geographic units remained in the same rurality quintile and <0.3% shifted by two or more quintiles across variants. Rank-difference analyses demonstrated that large changes were geographically limited and primarily concentrated in island areas. Convergent validity with physician distribution was moderate and consistent across variants (Spearman's ρ approximately -0.52). Criterion-related validity showed weak but significant negative correlations with life expectancy, which were stronger for males (ρ ≈ -0.27 to -0.29) than for females (ρ ≈ -0.17 to -0.19). These patterns were consistent with findings from the original RIJ study. We found no descriptive evidence that the associations with external indicators differed meaningfully across RIJ variants. The 95% CIs of the correlation coefficients largely overlapped across RIJ variants, suggesting no meaningful differences in their associations.

Despite differences in distance metrics and island classification, all RIJ variants captured highly similar underlying dimensions of rurality and demonstrated comparable validity. These findings indicate that rurality measurement in Japan is stable regardless of methodological refinements, allowing RIJ variants to be flexibly selected according to specific research or policy purposes without materially affecting conclusions.