PROTOCOL TITLE:

Natural experiments for translation in diabetes (NEXT-D)

PRINCIPAL INVESTIGATOR:

Principal Investigators:

Bernard S. Black, JD, MA

Chabraja Professor, Pritzker Law School and Kellogg School of Management

Northwestern University

Abel N. Kho, MD

Director, Institute for Public Health and Medicine (IPHAM) - Center for Health Information Partnerships

Associate Professor in Medicine-General Internal Medicine and Geriatrics and Preventive Medicine-Health and Biomedical Informatics

Feinberg School of Medicine

Northwestern University

Co-Investigators:

Laura J. Rasmussen-Torvik, PhD, MPH, FAHA

Assistant Professor, Department of Preventive Medicine

Center for Genetic Medicine

Feinberg School of Medicine

Northwestern University

John Meurer, MD, MBA

Director, Institute for Health and Society

Professor, Institute for Health and Society

The Medical College of Wisconsin

Russ Waitman, PhD

Director, Medical Informatics

Assistant Vice Chancellor, Enterprise Analytics

Associate Professor, Internal Medicine

University of Kansas Medical Center

Mei Liu, PhD

Assistant Professor, Internal Medicine, Division of Medical Informatics

University of Kansas Medical Center

VERSION NUMBER: 1.2

VERSION DATE: 5/25/2017


Table of Contents

1.0 Objectives 4

2.0 Background 4

3.0 Inclusion and Exclusion Criteria 5

4.0 Study-Wide Number of Subjects 6

5.0 Study-Wide Recruitment Methods 6

6.0 Multi-Site Research 6

7.0 Study Timelines 7

8.0 Study Endpoints 7

9.0 Procedures Involved 9

10.0 Data and Specimen Banking 11

11.0 Data and Specimen Management 12

12.0 Provisions to Monitor the Data to Ensure the Safety of Subjects 12

13.0 Withdrawal of Subjects 12

14.0 Risks to Subjects 12

15.0 Potential Benefits to Subjects 12

16.0 Vulnerable Populations 12

17.0 Community-Based Participatory Research 12

18.0 Sharing of Results with Subjects 12

19.0 Setting 12

20.0 Resources Available 13

21.0 Prior Approvals 13

22.0 Recruitment Methods 13

23.0 Local Number of Subjects 13

24.0 Confidentiality 13

25.0 Provisions to Protect the Privacy Interests of Subjects 13

26.0 Compensation for Research-Related Injury 14

27.0 Economic Burden to Subjects 14

28.0 Consent Process 14

29.0 Process to Document Consent in Writing 14

30.0 Drugs or Devices 14

1.0  Objectives

For the NEXT-D project we will study diabetes diagnosis, treatment, and outcomes in adults, using data from two “PCORnet” Clinical Data Research Networks (CDRNs), each of which contains electronic health records (EHR) from hospitals and other providers in a geographic region, with potential to expand to additional PCORnet sites. The primary objectives are as follows. To accomplish these goals, we have assembled a multidisciplinary team of researchers, including experts in EHR, diabetes, research design for causal inference, health economics, epidemiology, and law.

1.  To assess how Medicaid expansion affects near-term and medium-term diabetes diagnosis, treatment compliance, and health outcomes.

2.  To develop methods for combining difference-in-differences (DiD), matching, and multiple imputation.

3.  To assess diabetes-related economic benefits of Medicaid expansion.

2.0  Background

Diabetes is a highly prevalent, costly, and often undiagnosed disease. In 2012, an estimated 29 million Americans (9.3% of the US population) had Type diabetes; of these, nearly 30 % have not been diagnosed (CDC, 2014). Individuals with diabetes are at higher risk of several other diseases including heart attack and stroke, peripheral artery disease, diabetic retinopathy, and renal disease. In part because of these complications, average medical expenditures for people with diagnosed diabetes were 2.3 times higher than for people without diabetes in 2012. Estimated direct and indirect diabetes costs in the US in 2012 were $245 billion (CDC, 2014).

The American Diabetes Association has established guidelines for diabetes diagnosis and treatment. Individuals are diagnosed based on HbA1c and blood glucose levels (ADA, 2014). Individuals often do not exhibit symptoms for years after disease onset, which contributes to the high percentage of undiagnosed diabetes. Treatment effectiveness is monitored primarily by HbA1c testing, which should be at least twice per year in patients meeting treatment goals and quarterly for patients not meeting goals. A typical goal is an HbA1c less than 7%, although some providers may target 6.5% for healthier individuals or 8% for individuals with other comorbidities (ADA, 2015). For diagnosed type 2 diabetes, most patients should begin treatment with lifestyle changes, principally weight loss and exercise. If lifestyle changes alone do not achieve glycemic goals, oral metformin monotherapy should be initiated. If and when monotherapy does not achieve glycemic goals, combination therapy is indicated, then triple therapy, and finally combination injectable therapy involving injected insulin and other drugs. Because of the progressive nature of type 2 diabetes, most patients eventually need insulin therapy (ADA, 2015). In 2010-2012, among people 18 and older with diagnosed diabetes, 14% used insulin only, 15% used insulin and oral medication, 57% used oral medication only, and 14% used neither insulin nor oral medication (CDC, 2014).

Results from the Oregon Health Insurance Experiment showed increased diagnosis and use of diabetes medications but no decrease in HbA1c levels. This could reflect poor medication compliance (Baicker et al., 2013). In a recent study of the Medicare population, only 59% of diabetes patients were considered adherent to oral monotherapy (Tunceli, 2015).

The ACA strongly encouraged states to expand Medicaid eligibility to cover all persons up to 138% of the federal poverty limit (FPL), beginning Jan. 1, 2014. Federal funding would cover 100% of the additional cost for an initial period, and 90% after that. Persons with incomes above the 138% threshold would be eligible for subsidized insurance on state-organized health insurance exchanges. In practice, some states have thus far decided not to expand Medicaid at all, some have expanded in part but not up to 138% of FPL, some have delayed expansion, and some have negotiated state-specific waivers, covering aspects of their programs, with the Centers for Medicare and Medicaid Services, or are engaged in these negotiations.

Within our sample, Illinois, Indiana, Iowa, and Minnesota expanded Medicaid eligibility on Jan. 1, 2014 to 138% of FPL, Indiana expanded eligibility to 138% of FPL on Feb. 1, 2015, and Wisconsin expanded eligibility to 100% of FPL. The control states will be Kansas, Nebraska, Texas, and Missouri. This variation provides the natural experiment that we examine here.

Of central relevance to our research design, we have no reason to believe that state expansion decisions are either caused by, or associated with (except by accident), differences in the health of those who would become Medicaid-eligible. Instead, State expansion decisions have been heavily driven by local politics; liberal/Democratic states have generally expanded Medicaid; while conservative/Republican states often have not.

3.0  Inclusion and Exclusion Criteria

Our overall sample of persons for whom we will extract data from EHR repositories and geocode their addresses will be all persons with one or more encounter records in the EHRs for the CAPriCORN or Greater Plains Collaborative (GPC) institutions during our study period, from January 1, 2009 through December 31, 2016. Age eligibility range is 18-89 years old.

For many analyses, we will restrict the sample to a narrower group for which a diabetes diagnosis can be detected with reasonable reliability (if present). For example, a plausible narrower cohort is:

(i)  anyone with records for two office visits on or after January 1, 2009;

(ii)  Anyone with medical records consistent with a diabetes diagnosis (for example, a prescription for diabetes-related medication).

We will use the narrower “two visits” definition for many planned analyses, but will use the full sample to explore the sensitivity of results to alternate definitions.

Exclusion criteria are critical in defining the analytical sample. We plan to exclude persons with prescriptions for prednisone and other medications that may temporarily increase glucose levels. We also plan to exclude pregnant women for a period during and soon after pregnancy, as we do not wish to include gestational diabetes in our analyses.

The initial data extraction will cover the period from January 1, 2009 through December 31, 2016. We will conduct additional data extractions approximately every 6 months and update our protocol to reflect this as needed.

4.0  Study-Wide Number of Subjects

The potential study subjects which meet the sample selection criteria stated above, from the two CDRNs together are an estimated 25 million patients in five expansion states and four control states. The large sample will permit precise estimates of average treatment effects, and will let us assess whether treatment effects depend on various factors, including state- specific Medicaid policies, income, gender, and ethnicity.

The closest prior study, the Oregon Health Insurance Experiment, had available a much smaller sample and found that almost all predicted effects of Medicaid expansion were statistically insignificant. We require a very large sample in order to obtain reasonably precise estimates.

5.0  Study-Wide Recruitment Methods

Not Applicable.

6.0  Multi-Site Research

This study will be conducted at two Clinical Data Research Networks, CAPriCORN and the Greater Plains Collaborative, which in turn are comprised of multiple health care institutions. The overall study will be led by co-Principal Investigators Bernie Black and Abel Kho out of Northwestern University who will be responsible for the overall conduct of the study. The GPC participation will be overseen by site PI Russ Waitman. Progress will be coordinated through a monthly collaboration call between investigators at both CDRNs, led by the overall PIs.

Each CDRN will be expected to maintain an approved IRB study protocol and maintain full data security procedures in line with the policies in place at their CDRN and aligned with the policies and procedures of their own data contributing institutions.

Representatives from both CAPriCORN and GPC will participate in national committees of the overall Next-D consortium.

7.0  Study Timelines

There will not be any active participation of subjects in this study. Below find a proposed timeline for the project, assuming a January 1, 2016 start date. EHR data will be extracted for analysis through mid-2020, to provide as long a period as possible after the Jan. 1, 2014 expansion of Medicaid. Because it is anticipated that aspects of the EHR systems will be continually evolving over the grant, and because data will continue to accrue for many years of the grant, certain tasks are assumed to be repeated at near yearly intervals.

Project Year / Year 1 / Year 2 / Year 3 / Year 4 / Year 5
Project Quarter / 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10 / 11 / 12 / 13 / 14 / 15 / 16 / 17 / 18 / 19 / 20
Calendar Year / 2016 / 2017 / 2018 / 2019 / 2020
Task
Extract and harmonize EHR data from CAPriCORN and GPC sites
Conceptualize source cohorts, define prevalent and newly diagnosed diabetes
Define treatment and compliance phenotypes
Define diabetes outcomes phenotypes
Geocoding
Aim 1: Medicaid expansion and diabetes diagnosis, treatment, and outcomes
Aim 2: Develop and refine causal inference methods
Aim 3: Assess diabetes-related benefits of Medicaid expansion
Reports / Papers / Dissemination

8.0  Study Endpoints

1.  Develop methods for combining DiD, matching, and multiple imputation.

2.  Understand the health and financial effects of Medicaid expansion and possible policy implications on diabetes care and beyond.

3.  Conclude answers to research questions involving how Medicaid insurance affects diabetes diagnosis, treatment, patient compliance, and outcomes. Principal questions include:

i)  For the newly Medicaid-covered, how many were previously diagnosed with diabetes, how many are newly diagnosed, and how many remain undiagnosed, although likely to have diabetes or be near-diabetic?

Among those diagnosed with diabetes, impact on treatment and compliance:

ii)  How many take minimal treatment steps (obtain prescriptions for metformin and other standard medications, receive regular checkups, etc.)

iii)  How many effectively manage their diabetes (using as measures hemoglobin A1c (HbA1c) levels and a “diabetes bundle” which combines HbA1c, LDL cholesterol, and blood pressure levels)?

iv)  How often is screening compliant with guidelines for diabetes management (semiannual checkup, including HbA1c level; annual urine microalbumin, eye exam, and foot exam; immunization (pneumonia ever, flu annually)?

What are the medium-term changes in diabetes-related health outcomes for newly-insured Medicaid patients, including:

v)  renal disease and failure?

vi)  cardiovascular disease, including coronary events and ischemic stroke?

vii)  peripheral circulation and limb loss?

viii)  diabetic retinopathy?

Economic impact of diabetes treatment. For newly Medicaid covered diabetes patients:

ix)  what are the costs incurred by Medicaid, within the study period?

x)  what are the projected implications of improved patient health for longer-term costs, many of which will be incurred by Medicare?

xi)  what are the projected implications of improved patient health for long-term reduction in disability and increase in quality-adjusted life years?

Secondary health outcomes. We will also study several additional questions, but treat these as secondary goals due to data limitations or measurement challenges:

xii)  How often do newly Medicaid-insured diabetes patients fill their diabetes prescriptions?

xiii)  How many newly Medicaid-insured diabetes patients lose weight, including through bariatric surgery?

xiv)  How does Medicaid-coverage affect medium-term mortality (this requires linking patient records to state death records, which may be infeasible in some states)?

Subsample analyses and disparate impact:

xv)  How do any treatment effects vary with personal characteristics, including gender, ethnicity, education, employment, income, body-mass index, previously versus newly diagnosed diabetes, and diagnosed versus likely undiagnosed diabetes?

xvi)  Are treatment effects affected by various types of churn – persons moving in and out of insured status, or moving between Medicaid and private insurance, versus stably covered by Medicaid?

9.0  Procedures Involved

Measuring Diabetes Diagnosis, Treatment, and Outcomes

Extracting information from EHR: Our extensive experience with extracting information from a variety of EHR sources, through the eMERGE consortium and through CAPriCORN and GPC analyses, has demonstrated that algorithms for extracting EHR information need to be validated and often modified for the best performance, particularly when extracting information across multiple EHR sites (Kho et al., 2011). Below, we propose potential algorithms for extraction of various covariates, but these algorithms will likely be modified to some extent as we implement them. Further modifications will be needed beginning with the fourth quarter of 2015, when ICD-10 codes generally replace ICD-9 codes. Our previous experience in defining and modifying phenotypes and extracting lab values from EHR prepare us well to extract the information described below.