OHDSI Home | Forums | Wiki | Github

Are surveys measurements or observations?

CDM Builders
#1

Friends:

This question came up when we added LOINC Answer codes to the vocabulary as a response to Yifan’s request (CDM vocabulary - loinc answer code is not available). We didn’t have them before, but now we will.

The way the anser codes works is that for about 8600 LOINC codes (out of 74k) there are prescribed answer (value) codes. In Yifan’s exampe, in the MDS survey there is question 54583-0 ‘Federal OBRA reason for assessment’ with the possible answers

LA10-4 Admission assessment (required by day 14)
LA15-3 Quarterly review assessment
LA12-0 Annual assessment
LA13-8 Significant change in status assessment
LA10888-8 Significant correction to prior comprehensive assessment
LA10889-6 Significant correction to prior quarterly assessment
LA9-3 None of the abovef

(I am ignoring the quality of the answers, in particular the last one here.)

The way it would be in the CDM is that the concept_id contains the concept for the question, and the value_as_concept the answer code. But we could do that in either the OBSERVATION or the MEASUREMENT table.

The definition of a Measurement right now is: The MEASUREMENT table contains records of Measurement, i.e. structured values (numerical or categorical) obtained through systematic and standardized examination or testing of a Person or Person’s sample.

Is a survey a “structured and standardized examination or testing”? What do folks think?

Christian

Overhauling Measurement Types
Overhauling Measurement Types
#2

This question arises out of MDS 3 data so forgive me if I accidentally take the topic off on too specific a direction. I can’t speak about survey’s in general but we think there is a valid argument for handling SOME MDS 3 data as clinical observational data, SOME as conditions, SOME as measurement data, and SOME (but not much) in the procedure table. We think of MDS as a reporting procedure and we use the procedure table to store the various report types that you indicate in the background of your question (the other tables just don’t seem a good place for specifying reporting types). That MDS reports are reporting procedures, from our perspective, comes from the fact that they are carefully filled out by highly skilled registered nurses based on data directly from the a nursing home patient’s charts following a schedule mandated by Medicare. I am open to other suggestions though based how folks are handling similar data elsewhere.

Once you get past the report types, the range of health information present in an MDS report is quite wide and includes patient conditions (some recorded using ICD, others identifiable using LOINC question and answer codes), patient observations (behavior, falls, short stay status, etc), and measurements (depressive symtoms via PHQ-9, cognitive status using BIMS, etc). For our current ETL (CDM v4), we would like to use the LOINC question and answer codes in the observation table for all MDS observations and measurements, and code conditions separately in the condition table. But, when we migrate to v5, we would make some changes to use the measurement table for those things in the MDS that were derived using standardized surveys like the PHQ-9.

#3

For me it has to do with level of trust for responses. I trust measurements more than observations.

Measurements are performed by trained staff using specialty equipment and/or knowledge.
Observations are made by humans, often aided by technology but not necessarily.

“How many alcoholic drinks per week?”: Observation (likely inaccurate, too)
“Ethanol content in blood”: Measurement

When I think of surveys, I also think Likert scale. This concept is mapped to Observation.

Bill

#4

Rich:

I get your point. The domain depends on what the question is asking for.

Only problem with is approach: Somebody has to go in and make those determination. It is not impossible, but of course competes with all the other mapping and domain assignment activities.

If you didn’t get it your way, which of them would you prefer as a stop-gap?

C

#5

Bill:

I like that.

t