Ensuring the Quality of Data for Multi-Site Health Services Research - - PowerPoint PPT Presentation

Ensuring the Quality of Data for Multi-Site Health Services Research

Bradley G Hammill Duke School of Medicine & Duke Clinical Research Institute brad.hammill@duke.edu

Quality Across the Clinical Data Flow

Clinical Encounter Electronic Health Record Research Database Study- Specific Dataset

Many possible data quality intervention points

The PCORnet Experience

• Distributed Research Network

– 13 Clinical Data Research Networks (CDRNs) comprising 80+ sites – Use of Common Data Model (CDM) – Primarily electronic health record data – Control of data is local, not central – Queries are used to generate summary results for return

Clinical Encounter Electronic Health Record PCORnet Common Data Model Study- Specific Dataset

PCORnet Research Process

• Purpose: State needs for addressing study objective(s)

Step 1: Research Needs

• Purpose: Translate high-level clinical concepts to low(er)-level clinical concepts

to obtain from the data. Fill in any gaps posed by the research question. Step 2: Business Specifications

• Purpose: Provide detailed instructions for defining concepts based on the

PCORnet CDM Step 3: T echnical Specifications

• Purpose: Generate the SAS query to be sent to sites for execution

“PCORnet Query Programming Guidelines” Step 4: SAS Query

ADAPTABLE Trial

• Aspirin Dosing: A Patient-centric Trial Assessing Benefits and

Long-Term Effectiveness – Pragmatic clinical trial – Demonstration project of PCORnet – Leveraging EHR data – 20+ sites

• General query strategy

– 1-2 beta tests (limited sites) before official distribution “…designed to reflect ‘real-world’ medical care by recruiting broad populations of patients, embedding the trial into the usual healthcare setting, and leveraging data from health systems to produce results that can be readily used to improve patient care.”

Step 1: Research Needs

• Among the enrolled population, describe medical history and

prevalent conditions at baseline – Ex. Prior cardiac revascularization

• Among the enrolled population, summarize the rate of concurrent

medication usage, at baseline and throughout the course of the trial – Ex. Aldosterone antagonist

• Among the enrolled population, compare event rates between

treatment groups – Ex. Bleeding w/transfusion

Step 2: Business Specifications

• Define population

– Enrolled patients

• Define relevant time periods

– History: 1 year prior to enrollment – Follow-up: Up to 2.5 years following enrollment – Medication reporting: At baseline & every 6 months

• List of specific procedures that make up a concept

– Prior cardiac revascularization PCI, CABG, other? – Transfusion Whole blood, red blood cells, other?

Step 2: Business Specifications

• List specific drug names and ingredients for each medication

– Aldosterone antagonist Brand names: Inspra, Aldactone Ingredients: eplerenone, spironolactone

• List specific diagnoses that make up a concept

– Bleeding Intracranial hemorrhage Gastrointestinal hemorrhage Other?

• Other important things

– Medication usage Prescription or dispensing that covers a date

Step 3: T echnical Specifications

• Codes, codes, codes (+ some logic)
• Some things to keep in mind

– Do not make assumptions about the data (esp. based on your site’s data or experience with claims data) – Do specify comprehensive code lists – Do use validated algorithms where possible, but… – Do not limit yourself to validated algorithms – Do pre-test all queries – Do have a plan for site variability in data & results – Be flexible

Step 3: T echnical Specifications

Coding: Transfusions

• Specific transfusions (whole blood, red blood cells)

ICD-9-CM (Px)

99.03, 99.04

ICD-10-PCS

3023[0|3][H|N|P]1, 3024[0|3][H|N|P]1, 3025[0|3][H|N|P]1, 3026[0|3][H|N|P]1

HCPCS

P9010, P9011, P9016, P9021, P9022, P9038, P9039, P9040, P9051, P9054, P9057, P9058

• Other or non-specific transfusions

ICD-9-CM (Px) 99.0x (except above) ICD-10-PCS

[Many]

HCPCS

P90xx (except above)

CPT

36430

Issues:

• Study period crosses ICD-10

implementation date / 01-Oct-2015

• While at most sites other/non-spec

transfusions are ~25% of all transfusions, some sites are as high as 75%

• Annual transfusion rates in pre-test query

(general CV population) were about 3%? What to do with sites with <0.5%?

• What about revenue center codes?

Coding: Major Bleeding

• Specific diagnosis codes (too many to list)

ICD-9-CM (Dx) ICD-10-CM

• Additional logic

– Type of encounter: Inpatient – Diagnosis type: Primary – Timing: Between enrollment date & follow-up end date Issues:

• Study period crosses ICD-10

implementation date / 01-Oct-2015

• Annual major bleeding rates in pre-test

query (general CV population) were about 5%? No concerning site outliers.

• Validation studies have shown this to be a

less-than-reliably coded outcome

• Some sites in PCORnet do not have

primary diagnosis indicators for IP

• encounters. How to handle?

Coding: Aldosterone Antagonist

• Specific medication codes (too many to list)

Dispensing / National Drug Code (NDC) Prescribing / RxNorm concept unique identifier (RxCUI)

• Dispensing window

– DISPENSE_DATE > DISPENSE_DATE + DISPENSE_SUP

• Prescribing window (?)

– RX_ORDER_DATE > RX_END_DATE – RX_START_DATE > RX_END_DATE – RX_ORDER_DATE > RX_ORDER_DATE + RX_DAYS_SUPPLY – RX_START_DATE > RX_START_DATE + RX_DAYS_SUPPLY – RX_START_DATE | RX_ORDER_DATE in a defined period Issues:

• Use both medication tables? Not all sites

have both.

• Don’t forget to include discontinued codes.
• Which RxCUI term types to include?
• Need to know dose form?
• Need to know strength?
• How to handle missing end date & days

supply information in prescribing table?

Coding: Other issues

Study population – Trials = Enrolled – Observational = Loyalty cohort? Relevant time periods – Medical history look-back – “Current” lab values Procedures in the CDM – Sites have made many different decisions – Ex: Injected medications, E&M codes, any many more

Dealing with Data & Coding Variability

• Understand the diversity

– Data characterization results – Study-specific queries

• Have a plan

– Write algorithms and specifications “defensively” – Include multiple concept specifications

• Test queries, then re-test
• Acknowledge the reality of the data

– Potentially select sites based on initial query results – Show site-specific results as part of the report fit-for-use

/fit fawr yoos/ phrase Typically used to describe data that is capable of meeting specific study requirements Can also be used to describe sites that are capable

Summary

• Data quality requires planning
• Data quality results from attention to detail
• Data quality means acknowledging when data are less than perfect
• Data quality means dealing with data variability

THANK YOU! QUESTIONS?