In 2014, the state of Maryland partnered with the Centers for Medicare and Medicaid Services (CMS) to modernize its unique all-payer rate-setting system for hospital services to improve the overall health of Maryland residents by increasing health care quality and reducing the cost of care. In service of providing better care at lower costs, The Hilltop Institute at UMBC, in partnership with the Maryland Department of Health, has developed predictive risk stratification models to identify patients at high risk for potentially preventable health care utilization that can be used to help target care resources to the patients who need them most.
This document strives to explain the intended use, technical implementation, and model performance of the Hilltop Pre- Models as of December 2024. The Pre- Models are a suite of prediction tools spanning the Pre-AH Model, Pre-DC Model, and Pre-HE Model. This document will be updated as the models are updated or when new models become operational, and significant changes will be noted in the documentation edit history table and in the text when necessary.
Hilltop researchers Leigh Goetschius, PhD, Ruichen Sun, Fei Han, PhD, and Morgan Henderson, PhD, co-authored this article published in Health Services Research.
The emergence of algorithm-based health care models boasted the promise of objectivity since algorithms are theoretically free from the types of biases and errors to which humans are prone. In practice, however, data are not neutral, and these approaches can perpetuate biases and reinforce existing health disparities.
This study evaluates whether a large predictive model of avoidable hospital (AH) events was biased based on patient race or sex. This model assigns monthly risk scores to all Medicare fee-for-service (FFS) beneficiaries attributed to primary care providers that participate in the Maryland Primary Care Program (MDPCP). The researchers found no evidence of meaningful race- or sex-based bias in the model.
The Hilltop Pre- Models are risk prediction models developed by The Hilltop Institute at UMBC that use a variety of risk factors derived from Medicare claims data to estimate the event risk that a given patient incurs a given outcome in the near future. As of November 2022, there are three such prediction models in production for the Maryland Primary Care Program (MDPCP) population: the Hilltop Pre-AH Model™, which generates the “Avoidable Hospitalizations (PreAH)” scores; the Hilltop Pre-DC Model™, which generates the “Severe Diabetes Complications (Pre-DC)” scores; and the Hilltop Pre-HE Model™, which generates the “Hospice Eligibility and Advanced Care Planning (Pre-HE)” scores. These risk scores are displayed in the MDPCP Prediction Tools area on Chesapeake Regional Information System for our Patients (CRISP).
The Hilltop Pre-HE Model™—which generates the rankings for the Hospice Eligibility and Advanced Care Planning (Pre-HE) scores—is designed to support proactive advanced care planning discussions by estimating a patient’s risk of eligibility for hospice. The Pre-HE Model provides risk scores and reasons for risk for all attributed beneficiaries of Maryland Primary Care Program (MDPCP) practices every month in order to identify patients that are potentially appropriate for hospice care and to provide care teams with information that can guide the sensitive and difficult conversations about end-of-life care with patients and their families.
The Hilltop Pre-DC Model™—which generates the rankings for the Severe Diabetes Complications (Pre-DC) scores—is designed to facilitate the active management of type 2 diabetes by estimating individuals’ risk of incurring inpatient admissions or emergency department (ED) visits for severe diabetes complications. The Pre-DC Model provides risk scores and reasons for risk for all attributed beneficiaries of Maryland Primary Care Program (MDPCP) practices every month to help care teams proactively identify high-risk individuals and allocate scarce care management resources.
The Hilltop Pre-AH Model™—which generates the rankings for the Avoidable Hospitalizations (Pre-AH) scores—is designed to assist providers by allowing them to easily identify patients at a high risk of incurring an avoidable inpatient hospitalization or emergency department (ED) visit. The Pre-AH Model provides risk scores and reasons for risk for all attributed beneficiaries of Maryland Primary Care Program (MDPCP) practices every month to help care teams make informed decisions about how to direct scarce care coordination resources to the individuals who will benefit from them the most.
This annual report, written for the UMBC community, provides an overview of key projects and staff accomplishments for FY 2024.
Hilltop researchers Morgan Henderson, PhD, and Morgane Mouslim, DVM, SCM, published this article in the August 2024 issue of The American Journal of Managed Care. The first research of its kind, the piece compares the two different federally mandated sources of public, freely available health services price transparency data (insurer and hospital) for prices for maternity-related services negotiated between Blue Cross & Blue Shield of Mississippi and 26 Mississippi hospitals. Drs. Henderson and Mouslim examined the procedure code overlap for these pricing data sources, and then, for overlapping procedure codes, assessed price congruence. They found low levels of overlap: only 16.3% of hospital-billing code combinations appear in both data sources. However, for the overlapping observations, price concordance is high, with 77.4% of prices matching to the penny. The relatively low degree of overlap between the two pricing data sources indicates significant administrative misalignment between these pricing files; however, the strong concordance of overlapping prices suggests that these data sources are capturing pricing information from the same underlying contracts, as intended. This study is part of the ongoing research conducted by The Hilltop Institute on price transparency.
Hilltop researchers Morgan Henderson, PhD, Leigh Goetschius, PhD, and Fei Han, PhD, co-authored this article published in Medical Care.
Predictive models have proliferated in the health system in recent years and have been used to predict both health services utilization and medical outcomes. Less is known, however, on how these models function and how they might adapt to different contexts. The purpose of the current study is to shed light on the inner workings of a large-scale predictive model deployed in 2 distinct populations, with a particular emphasis on adaptability issues.
Senior Policy Analyst MaryAnn Mood, MA, gave a podium presentation at the 2024 AcademyHealth Annual Research Meeting (ARM) held June 29 – July 2 in Baltimore.