65 Predicting population-level vulnerability among pregnant women The permutation importance ranking (Appendix 2) yielded comparable results, although ‘mental healthcare utilization’ and ‘GP expenditures’ were ranked slightly higher. Appendix 2 additionally shows the rankings without self-reported data, using solely routinely collected data (31 variables). Figure 4. Variable importance ranking of the RF-model for ‘multidimensional vulnerability’. The set of 42 variables used for classification are ordered from high to low importance. The length of a line represents the importance of a particular variable on the model’s predictions. Self-reported data of the PHM-2016 is indicated with an asterisk (*). Results based on analyses among study population of 4172 women 3
RkJQdWJsaXNoZXIy MTk4NDMw