The Pregnancy Dataset We Can’t Afford to Lose

This is a love letter to a dataset: the Pregnancy Risk Assessment Monitoring Survey, or PRAMS. It is no secret that I love data, and part of that is an attachment to particular datasets that I believe give us valuable information or insights into the world. The PRAMS dataset is one of those.

My first goal is to explain why I love this dataset and try to give a sense of why ongoing public data collection of this type is so crucial for answering important policy questions.

I have a secondary motive, too. The PRAMS data is currently a casualty of a broader assault on data availability. Collection of this data is organized by the CDC, and, as of Spring 2025, it is no longer being collected. Even worse, researchers are no longer able to access historical data (although advocates have preserved some of the data and webpages). A huge amount of work has, for the moment, just vanished. It is not completely clear why this happened. Initially, the current administration stopped all projects that used terms like “equity” and “gender,” but it’s not clear why this data collection hasn’t restarted or why historical data is no longer being provided.

The disappearance of data doesn’t always get people animated the way other political moves do, and I think this is in part because people do not understand why it matters. So here’s why, for this one particular example.

What is PRAMS?

The Pregnancy Risk Assessment Monitoring System, or PRAMS, is a long-running survey run by the CDC together with state health departments. PRAMS started in the late 1980s, and it’s focused on learning about experiences before, during, and just after pregnancy. PRAMS is a survey — it doesn’t pull official medical records — which has some downsides (survey data always has limitations), but it allows the data to cover a lot of information that wouldn’t be captured in these other datasets.

The way the survey works is pretty straightforward. Each month, states draw a random sample of recent births from birth certificate files. Women in that sample get a survey in the mail, and if they don’t respond, the state will follow up by phone. This combination of mail and phone helps reach more people and makes the data more representative. States have an annual sample size between 1,000 and 3,000 women. 

PRAMS uses a core questionnaire that’s the same everywhere, with questions about things like prenatal care, health during pregnancy, and feeding choices for the baby. There is also some information drawn directly from birth certificates. On top of that, states can add extra modules from a standard list, or even write their own questions if there are issues they particularly care about. So the final dataset has both consistency, everyone is asked the basics, and flexibility: individual states can dig into topics that matter most locally.

Historically, researchers have been able to access the data — either individual states or the overall data — through an application to the CDC. They then use this data to analyze a whole host of topics, including prenatal care access, birth choices, postpartum depression, breastfeeding, and beyond.

How PRAMS data was used to study breastfeeding

Increasing breastfeeding rates is an important policy goal for many organizations. We have a general sense that breastfeeding rates in the U.S. are lower than in many peer countries, but that fact alone doesn’t explain why.

One way into this question is to learn more about the details of women’s breastfeeding journeys. For example, one reason for lower breastfeeding rates in the U.S. could be that many women never start breastfeeding; another reason is that they start but do not continue. If the goal is to increase these rates, then this distinction really matters. Better in-hospital support may increase breastfeeding initiation, but will not help continuation. Better support at home (paid leave, work accommodations) may increase continuation, but wouldn’t necessarily impact initiation.

The PRAMS is the best data source we have to answer this question, and it provides a lot of insight. In this paper, researchers used the data from 2016-2019 to look at breastfeeding duration and how it varied across educational groups in the U.S. The chart below shows breastfeeding estimates at birth, three, and six months of life by education group. 

The data pretty clearly shows a growing gap across groups over time. Relative to those with a high school degree, those with a college degree are 15 percentage points more likely to initiate breastfeeding but 30 percentage points more likely to still be doing it at three or six months. 

This suggests that a significant part of the issue is with breastfeeding continuation. One explanation for that is limited paid leave. How would one start to answer that question with data? Ideally, we’d have information on the amount of paid leave and breastfeeding duration.

It turns out that, in some of the PRAMS data, we also have that. A second paper uses these data to associate the length of maternity leave with breastfeeding duration. And, indeed, women who have shorter paid leave are also more likely to quit breastfeeding after a shorter duration of time.

Why losing datasets like PRAMS matters

The breastfeeding duration research I discussed above is just one small example of why these data are so useful.

A search in PubMed for uses of the PRAMS data reveals almost 1,000 published papers just in the journals they cover. These papers begin in the early 1990s (prenatal smoking behavior and birth outcomes, unintended pregnancies in Oklahoma) and continue through the present. The most recent paper listed in PubMed (September 20, 2025) uses the PRAMS data in a machine learning model to predict postpartum depression.

Papers using these data cover topics from dental care to prenatal visits to smoking cessation to breastfeeding. The PRAMS covers data on intimate partner violence. It lets us look at where maternity care is more accessible and where it is less, and why that might matter.

Many of the papers using the PRAMS data are, themselves, sources for other papers. A paper on the prevalence of gestational diabetes in the 2007 – 2010 period has over 800 citations in Google Scholar. Meaning over 800 other researchers found this result important enough to influence their own work.

What I think is notable about the PRAMS — similar to many other large public health datasets like the National Health and Nutrition Examination Survey or the National Health Interview Survey — is that it serves at least two broad purposes. First, there is a lot of research that just uses these data as the central source. Second, there are a lot of papers where this is a supportive dataset. Researchers may use it to argue that their main dataset is broadly representative. They may merge their data with the PRAMS to have a broader or larger sample. This usage is only really possible because the data is public and easy to access; it’s a benchmark that a broad set of researchers can use and understand.

Because data collection for the PRAMS has shut down and the CDC has stopped processing applications for access to the existing data, this means, fundamentally, the data is lost. At least for the moment. (It might come back, but it’s hard to know.) We will know less about the experiences of pregnancy and early parenting and have less ability to understand what influences them positively and negatively. We will simply know less that can help us make people’s lives better.

If you’d like to do something about PRAMS and the broader issue of data loss, please share your story with the Data Foundation on why this dataset matters to you.

The bottom line

• The Pregnancy Risk Assessment Monitoring System (PRAMS) is a long-running survey run by the CDC that gathers detailed, state-level information on women’s experiences before, during, and after pregnancy.
• Historically, researchers have been able to access this data to study a wide variety of topics, including understanding why the U.S. has lower breastfeeding rates than peer countries.
• Over the past three decades, PRAMS data has been used in nearly a thousand studies (including breastfeeding duration), serving as both a primary and benchmark dataset for public health research.
• With data collection now halted and access restricted, we risk losing one of our most valuable tools for understanding and improving pregnancy and early parenting experiences.

Get more from ParentData when you sign up

This article is part of your free preview. By signing up, you can get:

Access to ParentData

Newsletters delivered to your inbox

Community discussions with other data-driven parents

Sign up

Community Guidelines