How We Tracked Workforce Reductions at Federal Health Agencies

The Trump administration has refused to reveal how many workers have left federal health agencies amid a massive purge. Without official figures, ProPublica turned to a federal employee directory to quantify the impact.

Several news outlets — including ProPublica — have used this directory to show who is entering and leaving the federal government, spot political appointments and identify members of the Department of Government Efficiency. According to multiple former and current employees and health agency documents, the HHS employee directory helps workers locate and authenticate their colleagues. HHS did not respond to ProPublica’s questions about why it wouldn’t share data on workforce reductions.

To understand staffing changes over time, ProPublica has been regularly archiving the HHS directory since before President Donald Trump returned to office. Unlike official government worker datasets, which are often months out of date and incomplete, directory data provides a more current picture of who is, and isn’t, employed at the nation’s largest health agencies.

The HHS directory covers staff in the department’s main office and across more than a dozen health agencies and institutions, including the Food and Drug Administration, the Centers for Disease Control and Prevention and the National Institutes of Health.

The directory provides a worker’s name, email address, agency, office and job title. Some employees are flagged as “non-government,” indicating they may be working on contract or have another temporary status.

While official HHS employment figures suggest its workforce is about 82,000 employees, the directory contained nearly 140,000 entries. The difference is explained by nongovernment workers such as contractors, fellows, interns and guest researchers. Roughly 30,000 records in the directory were explicitly marked as nongovernment, but this label was far from comprehensive. We found hundreds of workers who had titles that suggested they were contractors but were not flagged as nongovernment.

While these workers were not directly employed by the government as civil servants, we included them in our analysis because they are crucial to agency operations, particularly at research-heavy institutions such as the NIH, FDA and CDC. We excluded interns, students and volunteers, as well as directory entries that were tied to group mailboxes or conference rooms rather than individual people.

To analyze turnover, we tracked when a worker’s email address first appeared and last appeared. When an email address disappeared, we marked it as a departure. When a new one appeared, we marked it as a hire. To quantify cuts to the workforce, we counted the number of entries that appeared in the directory before Jan. 25 and disappeared on or after that date. We chose Jan. 25 to account for delays we observed in updates to the directory. Starting the analysis a few days after Trump was sworn in should exclude most political appointees who left at the end of the Biden administration from our analysis, but there may be cases where a political appointee was dropped from the directory after this date. Our analysis includes departures through Aug. 16.

While our analysis is intended to understand workforce cuts, not all departures are layoffs. Workers may have left for other reasons, including retirements, resignations, buyouts and contract cuts.

To test the directory’s accuracy, we spot-checked employment status using LinkedIn, other open-source records and interviews. While we did not find up-to-date, publicly available profiles that matched the directory for every person we searched, we were able to confirm the employment status for dozens of current and former workers.

Some high-profile departures enabled us to test our methodology: Dr. Vinay Prasad, who was appointed to lead the FDA’s Center for Biologics Evaluation and Research in May, left the agency in July and resumed his role again in August. The directory data accurately reflected each of these employment changes.

We also wanted to understand the role of the lost workers and how their departures would affect their agencies. For this, we relied on an “organization” field in the directory, which allowed us to link employees to their specific office within HHS using public organization charts. At large agencies, this field proved incredibly valuable to understanding the exact divisions where the losses took place.

Job titles in the directory are inconsistent, so we used a local large language model classifier to assign each title into one of four major groups: science/health, regulators/compliance, tech/IT and other. We manually reviewed the job titles in each category and looked up former employees on LinkedIn to confirm their employment status and nature of the work they performed. Our totals for the NIH are an undercount because 78% of entries there do not list a role. We relied on workers’ most recently listed role, which may reflect a job a person had for only a short time.

The employee directory is the best data we have access to, but it is not without limitations.

Our understanding is that each record represents an individual worker. There are some cases where our analysis counts a worker twice because they were seen in two agencies or had multiple email addresses, likely due to a name change or clerical fix. Our review of these records suggests these types of duplicates are uncommon.

Our rigorous review found some instances where the directory is not up to date. We excluded the Centers for Medicare & Medicaid Services from our analysis because its directory has not received regular updates. We also excluded the Indian Health Service because it has an abnormally high rate of churn.

We are not counting about 2,500 entries that can be found in search but that are missing a detailed record or do not list an email.