What do we mean by “data-driven” pregnancy care?

Isabel Fulcher
Chief Scientific Officer
at Delfina
min read

In digital health, we often hear general terms like “data-driven decision making” and “predictive analytics” – but what do these actually mean in practice? In this blog post, I would like to describe how we define “data-driven” pregnancy care at Delfina and why we believe a data-driven approach improves patient experiences and outcomes. 

A key component of any data-driven program is—to no one’s surprise—data! We invest heavily in our data systems to ensure that we are collecting quality data for every participant. And what makes data quality data? To us, quality data is information that accurately reflects a patient’s past and current health statuses in addition to information that may have an impact on health status, such as family history of illness or a patient’s access to housing, food, or education. To build a comprehensive pregnancy story for each patient, we collect data from three sources: electronic health records, remote patient monitoring, and patient-entered sociodemographic information. As we have previously described, simply having access to these data sources does not necessarily correspond to “quality data” as missing information within these systems may not accurately reflect a patient’s health status.

We use these data sources to deliver data-driven pregnancy care with three overarching outputs: (1) delivering tailored care, (2) enabling timely decision-making, (3) serving diverse patient needs (Figure 1). 

Figure 1. Data sources and outputs for data-driven pregnancy care at Delfina 

(1) Delivering tailored care

The data science team develops machine learning models to identify patients that may benefit from interventional workflows. To ensure these models can be used to deliver tailored care to the patient’s benefit, we work with our clinical team to identify the appropriate time window for rendering model results. For example, prophylactic aspirin use in the first trimester has been shown to reduce the risk of preeclampsia by 10-20%. Thus, our model for hypertensive disorders of pregnancy only uses patient information prior to the end of the first trimester to identify patients who may benefit from prophylactic aspirin use (Figure 2).  We employed a similar approach for gestational diabetes in our recent study on machine learning for gestational diabetes.

A lot of work goes into model development, but it is only the first step to delivering personalized care! Crucially, we need to communicate model results in a way that is interpretable and actionable to providers. This requires the collective brain power of our data science, clinical, product, and engineering teams to develop a user interface that not only renders the model results, but also enables the provider to determine if a patient may benefit from specific interventions. Then, the provider can initiate those interventions! 

Figure 2. An example of delivering tailored care: mitigating the development and escalation of hypertensive disorders of pregnancy

(2) Enabling timely decision making

Our platform includes remote patient monitoring for biometric measures, such as mood, symptom, weight, blood pressure (if indicated), and blood sugar (if indicated) throughout their pregnancy. Compared to the standard of care, we are collecting significantly more biometric measures, which can enable timely decision making from the patient, provider, and our Delfina Guides (Figure 3). Specifically, we have developed data visualizations for our users that communicate the relevant aspects of a patient’s pregnancy journey. For patients, we display their inputted biometric information, so that patients can view their trends over time. For providers, we utilize provider-set thresholds to highlight patients for review so that providers can identify changing trends for specific patients. For Delfina Guides, we display patient engagement, so that they can follow-up with patients and assess their specific needs. 

Figure 3. Enabling timely decision making with Delfina Care (white and gray dots represent standard in-clinic visits and purple dots represent biometric entries) 

(3) Serving diverse patient needs

In order to meet patients where they are, we monitor and evaluate our Delfina Care program using data collected as a byproduct of implementation. That is, we track program engagement and health outcomes by patient groups, such as race and ethnicity, language, age, prior pregnancy history, and insurance type, to ensure that certain populations of patients are not differentially served or impacted by Delfina Care. When gaps are identified, we work with our Delfina Guides to develop new programs that improve those aspects of the care experience. For example, we recently started “Mom-To-Be Support Groups” to foster a sense of community for pregnant patients living in similar geographic areas. After new programs or technologies are deployed, we evaluate, modify, and repeat the process! 

At Delfina, we believe that a data-driven care system can uniquely improve health outcomes. By delivering novel insights from quality data, we can transform the level of practice for providers and the experience of care for patients. 

Delfina on social media
© 2024 Delfina
info@delfina.com · 2021 Fillmore St, Ste 37 · San Francisco, CA 94115