66 CHAPTER 3 DISCUSSION This study provides insight into predicting multidimensional vulnerability during pregnancy at population-level in the Netherlands using pre-pregnancy routinely collected data and the relevance of additional self-reported data on health, wellbeing, and lifestyle. Based on our results, it seems reasonably feasible to predict multidimensional vulnerability using solely routinely collected data, since the RF-model could distinguish between those with and without multidimensional vulnerability and was able to correctly predict multidimensional vulnerability in many cases. However, we found that adding self-reported data improved model performance. Out of the seven strongest predictors to multidimensional vulnerability in our dataset, three concerned self-reported health, two concerned socioeconomic characteristics, and two related to healthcare expenditures and utilization. Using solely routinely collected data to predict multidimensional vulnerability appears feasible, but several women were wrongly assigned to the vulnerability class, and other cases were missed. The crucial concern is whether the model achieved adequate performance, prompting consideration of using this readily available routinely collected data versus acquiring self-reported data on experienced health. Both data sources have advantages and disadvantages, and may be used for different purposes. Using routinely collected data is relatively easy, accessible and time efficient. This pragmatic approach recognizes that not all data are available and can be utilized, analysed and interpreted. However, it is less accurate which might mainly affect those missed by the model. Considering all relevant factors by using additional self-reported data leads to better predictions. However, this has numerous implications and inherent challenges, including increased burden to practitioners, time and costs. Based on our study, we consider routinely collected data sufficient for policy monitoring of multidimensional vulnerability at population-level. It can offer insight into its scope and development over the years and help identify municipalities and neighbourhoods characterized by increased vulnerability, enabling tailored (preventive) measures for efficient budget allocation. Simultaneously, we agree with previous scholars that applying vulnerability in a dichotomous way is challenging as the concept is multi-layered, contextualized and dynamic, requiring caution to avoid over-inclusion or exclusion of individuals (33, 34). Our previous study (12) revealed a greater array of vulnerability groups, with women having risk factors within one specific domain and protective factors in others. We must not overlook these and other intermediary and personal, contextual forms of vulnerability. Our predictive RFmodel was not intended for application in individual predictions and individual decisionmaking but meant for risk-stratification on a population-level. Because risk assessment is not straightforward, we consider routinely collected data by itself unsuitable for individual predictions, given that it insufficiently accounts for protective factors and coping strategies at an individual level, among others. We believe that an open conversation with (future) parents about their experienced health and well-being is indispensable to better understand their context and needs. It is essential that this is accompanied by a trusting relationship, and appropriate follow-up steps, preventing stigmatization, simplification
RkJQdWJsaXNoZXIy MTk4NDMw