Best Practices for Interoperable Data Exchange Using LOINC

advertisement

Best Practices for Interoperable

Data Exchange Using LOINC

Ming-Chin (Mark) Lin, MD

Stanley M. Huff, MD

Introduction

• Two primary use cases

– Sharing data between and among different institutions for patient care

– Aggregating data between and among different institutions for clinical research, quality improvement, public health surveillance, etc.

(secondary use)

• Use LOINC codes as the lingua franca for the data sharing

Introduction (continued)

• Mark’s work comparing LOINC usage across ARUP,

Regenstrief, and Intermountain

• What is truth?

– If local codes from different sites are mapped to the same LOINC code, how do we know they are really the same test?

– If local codes from different sites are mapped to different LOINC codes, how do we know they are really different tests?

• Extensional definitions

– Comparison of names (substance, timing, property, specimen), units of measure, mean value, standard deviation, coded values, co-occurring tests, etc.

• Results: We found about a 4% error rate in mapping

– And that is us! What is it like for “regular” facilities?

Introduction (continued)

• Analyzing the errors lead to additional questions

– Can we classify the errors?

– What is the ultimate goal of mapping?

– Can we define “best practices” for mapping so that everyone doing mapping can achieve greater accuracy?

“Fit for Purpose” or “Good Enough” mapping

• Example: Tests with a method specified at site

A are mapped to methodless tests at site B

• Works for the known use case

– Either estimated weights or scale weights may be good enough for a particular study

• This represents a loss of information when data moves from A to B

Proposed Best Strategy

• Always map to the LOINC code or combination of codes that capture all known information about the test

– Always capture the method in the data if it is known

• Rationale: All uses of the data (especially secondary uses) are not known at the time of initial mapping or data collection

• “Fit for Purpose” mappings will preclude secondary use of the data in some situations

– What if you want to study whether two different test methods are truly equivalent?

Degrees of Interoperability

• Degree I: Exact equivalence without translation

– Same code, unit of measure, and value set

– Data are mutually substitutable in all contexts of use

• Degree II: Exact equivalence after translation

– Unit of measure conversion (need UCUM)

– Mass concentration to substance concentration conversion

(need the molecular weight)

– Pre and post coordination translation

• Method as part of LOINC code versus method sent somewhere else in the message

• Peak or trough as part of LOINC code versus peak and trough sent somewhere else in the message

– Data are mutually substitutable in all contexts of use after translation

Degrees of Interoperability (cont)

• Degree III: Context specific subsumption

– A parent-child relationship exists between tests at the different institutions

• Method specific tests roll up to methodless tests

• IgM or IgG antibodies roll up to generic antibody

– Data are mutually substitutable only in a specific

context of use even after translation

• Degree IV: No interoperability

– No comparable data or information exists between or among institutions

Examples

Proposal

• Create specific best practice mapping guidelines for difficult situations and common errors

• Examples

– How to deal with variable granularity in methods

– How to deal with pre and post coordinated specimen type

– How to deal with pre and post coordinated challenge conditions

– Use of Acnc and Titr

Example Guideline for Method

• If possible, and the method is known, map to the methodless LOINC code and always send the method in some other part of the message

– Related policy: the LOINC committee will make all needed methodless LOINC codes

• If the method is known but it is not possible to post coordinate the method, map to the method specific pre coordinated LOINC code

• If the method is not known, map to the methodless LOINC code

Example Guideline for Interpretations

• Always map to the quantitative LOINC code

– Related policy: The LOINC Committee will discourage or deprecate the use of nominal or ordinal LOINC codes for concentrations

• Send numbers when they exist as the value of

OBX 5

• Send interpretations when they exist as the value of OBX 8

• One or the other or both of the numeric value and the interpretation can exist in a data instance

Discussion

Download