Finding Hidden Patterns in Health Data Using Data Mining Technology

Public health nurses address serious health and social problems with low income, high risk families. The link between poverty with its related risk factors and poor health outcomes has been well established. Children of clients experiencing poverty, discrimination, lack of social support, and early parenting show a disproportionately high incidence of insecure client-infant attachment as well as poor health and social outcomes. It is generally accepted as an effective approach to improving life course trajectories for low income, first time mothers and their infants.  However, variations in outcomes by population and program have been noted in the literature, suggesting that there is a need to improve the quality of home visiting care. Given that some high risk mothers and children can benefit from home visiting care, and resources for care are limited, there is a critical need for realistic clinical decision support based on real world data for these complex clients and situations. To date, home visiting care outcomes research has primarily been conducted from a black box perspective; that is, the intervention is a nurse-visited condition, without detailed analysis of the visit or intervention content, and employs classical methodological and statistical techniques that control for variation in the sample. In this study, we used data mining techniques with Omaha System data, and clustering family home visiting data revealed patterns for four unique client profiles (depressed, unhappy parents; low income, first time pregnant parents; very high risk parents; and short-term postpartum parents) and 14 intervention clusters that addressed diverse problems with multiple actions. We then employed a binomial generalized estimating equations model to look for significant associations between profiles, interventions, and outcomes. We found that low income, first time pregnant parents showed significant associations between different intervention clusters and low vs. high outcomes.