88101964F102.jpg

Laura A. Hatfield, PhD

Dr. Hatfield is an Associate Professor of Health Care Policy (Biostatistics) at Harvard Medical School's Department of Health Care Policy. Her methods research centers on trade-offs among multiple outcomes, with an emphasis on hierarchical Bayesian modeling. With Dr. Sherri Rose, she co-leads the Health Policy Data Science Lab and the Methods Core of the Healthcare Markets and Regulation Lab, funded by the Laura and John Arnold Foundation. Dr. Hatfield's research in the Methods Core focuses on difference-in-difference studies.

Recent honors include the 2018 ISPOR Health Economics and Outcomes Research - Methodology award for her paper on incorporating loss functions in safety surveillance decisions. Dr. Hatfield is the 2019 Chair-Elect of the Health Policy Statistics Section of the American Statistical Association and is serving a three-year term on ENAR's Regional Committee (RECOM). She previously served on the Editorial Board of Medical Decision Making. 

Dr. Hatfield earned her BS in genetics from Iowa State University and her PhD in biostatistics from the University of Minnesota.

did_conceptual.png

Difference-in-differences

To evaluate an intervention that was not randomized, we can compare how groups affected by the intervention change and compare this to unaffected groups. This controlled pre-post design is known as a difference-in-differences study. To make causal conclusions, we must believe that the control group's change represents what would have happened in the intervention group had the intervention not occurred.

+ Papers and talks

Overview of methods ideas [slides]

Matching on time-varying covariates can lead to bias when treated and control come from different populations [paper | thread], but fix bias when treated units are selected on transient differences [commentary | reply]

Testing for parallel pre-trends: passing is not as good as you think and failing is not as bad as you think [preprint | slides]

loss_functions.png

Loss functions for decision-making

How can patients and doctors more effectively weigh the risks and benefits of treatment, given an individual patient's preferences? How can regulators best respond to evidence of emerging safety signals, given their policy priorities? By combining loss functions that express the costs and benefits of each possible action with Bayesian joint models for multiple uncertain outcomes, we can obtain optimal decisions in real-world settings.

+ Papers and talks

Patients and physicians can use loss functions to trade off risks and benefits of treatment options [paper | slides]

Regulators can use loss functions to respond to safety signals [paper | slides] in post-market surveillance [perspective]

Medicare beneficiaries choosing supplemental insurance can integrate financial and health concerns using utility-based decision-making [slides]

hierarchical_variance.png

Understanding variance: clustering and hierarchical modeling

Characterizing and communicating variation is difficult in many applied settings. This research explores two avenues for reducing complex patterns into interpretable summaries. 

+ Papers and talks

Clustering longitudinal trajectories shows variation in patient experiences [slides]

Linear combinations of Kronecker products allow flexible dimension reduction for "medium-dimensional" hierarchical models [paper | slides]; using hierarchical models of health plan quality yields personalized ratings [paper | slides]

imputation.png

Joint models for mixed outcomes

Many outcome types are used to quantify the evolution of a disease process. For example, patients may report their symptoms at regular clinic visits, and the timing of critical clinical events recorded. Bayesian joint models can improve bias and efficiency and handle missing data.

+ Papers and talks

Joint models for survival times and zero-inflated longitudinal outcomes [paper] and multiple longitudinal outcomes [paper | slides]; graphical displays for joint model results [paper]

Model averaging improves dynamic predictions from joint longitudinal-survival models [paper | slides]

Joint models can improve inference when outcomes of mixed types are missing in meta-analyses [paper | slides] and improve treatment effect estimates when one outcome makes up for lack of information in the other [paper | slides]

price_transparency.png

Program and policy evaluations

Payers and providers and constantly innovating in pursuit of the triple aim of improving health outcomes, quality of care, and health care spending. These interventions are rarely randomized, so evaluations rely on quasi-experimental designs, especially difference-in-differences. This research evalautes price transparency, accountable care organizations, patient-centered medical homes, global budgets, and telehealth.

+ Papers and talks

Medicare Shared Savings Program accountable care organizations (ACOs) demonstrate small spending decreases, concentrated in independent primary care groups (versus hospital-integrated groups), that grow over time [paper]

Hospital global budgets in Maryland: little evidence of utilization changes in urban [paper] or rural [paper] programs

A patient-centered medical home with financial incentives showed little evidence of impacts on utilization or spending [paper]

Ongoing randomized trial of telephone-based care coordination to reduce hospitalizations for home care recipients is in progress [paper]

Physician-facing price transparency at ordering makes little difference in spending or utilization [adult paper | [peds paper]; consumer-facing price transparency tools are rarely used [paper] and did not reduce spending among California public employees and retirees [paper] and two large employers [paper]

microsim.png

Documenting variation

To identify policy targets and evaluate the success of interventions, we need evidence on the sources of variation in important outcomes. This research documents variation in health care spending, utilization, outcomes, and coverage at patient, provider, facility, and geographic levels.

+ Papers and talks

Only 1/3 of patients with mental health care needs have substantive mental health discussions with providers at wellness visits [paper]

Medicare beneficiaries face high out-of-pocket health care spending that is projected to grow and especially impacts near-poor seniors [paper | slides]

Families of patients with cancer report better outcomes with hospice [paper]; older women with ovarian cancer receive intensive end-of-life care despite high hospice enrollment [paper]

Hospitalized patients cared for by their own PCPs versus hospitalists stay slightly longer, but are more likely to be discharged home and less likely to die within 30 days after discharge [paper]

In the US, there is large variation in the patterns of care for patients with advanced cancer [paper | slides], treatment for depression in children and adolescents [paper], and health insurance coverage gaps around pregnancy and delivery [paper]

We also see large variation in use of implantable cardiac electric devices to treat heart failure [paper]. Current electrical leads for implantable devices rarely fail [paper | meta-analysis]; evidence is still needed on choosing the best treatment in implantable device therapy for heart failure [paper]