63 Predicting population-level vulnerability among pregnant women Figure 2. Heatmap visualizing the geographical distribution of multidimensional vulnerability during pregnancy in the Netherlands, at municipality level, for all pregnancies from 2017 to 2021. A darker color indicates a higher percentage of vulnerability. Results based on analyses among all unique pregnancies from 2017 – 2021 (n = 807.904) Adding self-reported data to predictions The baseline F1-measure (using routinely collected data; 31 variables) was 0.70 and the potential optimum (using both routinely collected data and all self-reported data of the PHM-2016; 42 variables) was found to be 0.83, shown as vertical lines in Figure 3. Including self-reported variables improved the performance of the RF-models with solely routinely collected data. Especially self-reported data on ‘perceived health status’ (average 0.80) and ‘restricted by health’ (0.79) improved the model’s performance, but also ‘long-term illness’ (0.77) and ‘risk to depression or anxiety disorders’ (0.74). Others had little impact or slightly decreased performance, such as physical activity. Appendix 2 presents the results of adding two varying self-reported variables. This further improved the performance of the model. 3
RkJQdWJsaXNoZXIy MTk4NDMw