Safti net overview ahrq stakeholders mtg oct 2011

+

SAFTINet: Scalable Architecture for Federated Translational Inquiries Network

October 2011

+

Outline Overview

CER Objectives

+SAFTINet Overview and Aims

Funding Mechanism: AHRQ ARRA OS: Recovery Act 2009: Scalable Distributed Research Networks for Comparative Effectiveness Research (R01)

To build the infrastructure for a distributed data network that supports comparative effectiveness research

Intended components of the network

EHR and Medicaid claims data harmonized to a common data model (modifications to OMOP)

Made available for distributed query using TRIAD grid technology

Supplemented with PROs and practice-level survey data

+

SAFTINet Research Objectives

Develop four cohorts of patients with asthma (childhood and adult)hypertensionhypercholesterolemia

Conduct comparative effectiveness research on healthcare delivery system factors as they relate to health outcomes in these cohorts

+CER Hypothesis

Health care delivery system factors, such as the patient-centered medical home…

DELIVERY SYSTEM FACTORS + COVARIATES →

OUTCOMES(chronic disease

control)

+CER Hypothesis

Health care delivery system factors, such as the patient-centered medical home…

are important determinants of the control of asthma, high blood pressure and hypercholesterolemia.

DELIVERY SYSTEM FACTORS + COVARIATES →

OUTCOMES(chronic disease

control)

+Building infrastructure for CER

ConsiderationsAre we collecting the right data to support CER?Are the data of sufficient quality to conduct high

quality CER?Do we have sufficient power to detect significant

effects?Have we identified the right covariates to control

for bias and confounding?

+Methods and analysis

Example hypothesis: Asthma outcomes among adults are better at health

centers that implement PCMH functions

Unit of analysis? Outcomes are measured at the patient level Predictors are measured at the practice level Covariates (mediators, confounders, etc) exist at multiple

levels (patient, provider, practice, organization)

Analytic strategy: Hierarchical linear models (aka mixed effects or multilevel models) Can handle multiple levels of analysis in a single regression

equation Can handle missing data (requires at least two data points

per unit of analysis)

+Challenges

Limitations in use of “real world” clinical data for research purposes Variability in documentation across providers and systems E.g., ICD-9 code may not mean a patient HAS that diagnosis

– may represent a “rule out” or “considered” code Differences across systems and practices in the collection

of patient-reported outcomes data (point of care vs non point of care)

Sample size and power Highest-level unit of analysis?

Organizations (n = 5) Practices (n = 50) Providers (n = 190) Patients (n = 440,000)

+Challenges: Level of Analysis

At what level(s) do we measure each variable? Is PCMH a factor that varies at the clinic level or

the organizational level? How do we link patients to a provider or

practice?

How do we structure our analytic plan? What are the potential confounders of the

relationship between PCMH and disease outcomes?

How do we minimize the number of covariates given the limited degrees of freedom?

What biases in the data do we expect (and at what level of analysis)?

+Challenges: Level of Analysis

LEVEL DELIVERY SYSTEM FACTORS COVARIATES

OUTCOMES(chronic disease

control)

Organization

Practice

Provider

Patient

+Challenges: Level of Analysis

LEVEL DELIVERY SYSTEM FACTORS COVARIATES

OUTCOMES(chronic disease

control)

Organization

Practice

Provider

Patient

Partner organizations report PCMH is implemented at the organization level

+Challenges: Level of Analysis

LEVEL DELIVERY SYSTEM FACTORS COVARIATES

OUTCOMES(chronic disease

control)

Organization

Practice

Provider

Patient

Partner organizations report PCMH is implemented at the organization level

PCMH survey asks about how providers use the elements of PCMH in clinical care

+Challenges: Level of Analysis

LEVEL DELIVERY SYSTEM FACTORS COVARIATES

OUTCOMES(chronic disease

control)

Organization

Practice

Provider

Patient

Partner organizations report PCMH is implemented at the organization level

PCMH survey asks about how providers use the elements of PCMH in clinical care

The “PC” is about patients—should we ask for their input or assess their behavior?

+Challenges: Confounding

In real-world setting, different practices measure variables differently establish minimum requirements, e.g., for

implementing and reporting data from a PRO

How do we address common-cause variables? Use of Directed Acyclic Graphs (DAGs) to identify

a minimal set of covariates to remove confounding

+Challenges: Confounding

LEVEL DELIVERY SYSTEM FACTORS COVARIATES

OUTCOMES(chronic disease

control)

Organization

Practice

Provider

Patient

A practice’s degree of PCMH-ness is likely associated with how it implements a PRO and with the quantity and quality of data it reports for a PRO

+Questions/Comments/Input

“Knowing is not enough; we must apply. Willing is not enough; we must do.” —Goethe