OHDSI Home | Forums | Wiki | Github

Mammography codes

(Kristin Kostka, MPH) #1

@Dymshyts - the Women of OHDSI are looking for your expertise on procedure codes for mamography. Can you take a look at our T cohort: http://www.ohdsi.org/web/atlas/#/cohortdefinition/1771330?

We need to check the set of mammographies are accurately coded.

Appreciate your help!

(Anna Ostropolets) #2

@Dymshyts just to give you some additional background : Rupa and I identify the codes that are used for billing purposes. We decided to go with billing codes only as they give good coverage and are believed to be accurate. In other words, we didn’t include other broad concepts from SNOMED.

(Christian Reich) #3

Which will limit it to US-based data assets. Is that the idea?

(Anna Ostropolets) #4

Yes, looks like that.

(Kristin Kostka, MPH) #5

@Dymshyts We still need your help taking a look at the procedure codes we selected for mammography.

There’s two cohorts:

Can you take a quick look? We’re trying to button up the package and start running ASAP. Thank you!!

(Dmytry Dymshyts) #6

I pass the ball to @Eduard_Korchmar as he’s more an expert in procedure domain. :slight_smile:


But why are there SNOMED concepts?

(Eduard Korchmar) #7

Hello! I’ve taken a look at concept codes selected for “Screening mammography” and noticed that no codes from ICD-10-PCS vocabulary. Of course, they may have been excluded on purpose, since it is impossible to differentiate between screening mammography and any other reason why the procedure could have been performed, but I still ask just in case.

Also, SNOMED’s concept 4147961 “Screening for malignant neoplasm of breast” has descendants that are imaging procedures, but not radiographies; it is also possible that non-imaging types of screening (like blood tests for genetic markers?) could have custom mappings to this concept.

Do you need to include all imaging procedures on breast that could be screening procedures or are you looking specifically for X-Rays of breast?

(Anna Ostropolets) #8

@Dymshyts Broad = mammography without specifying that it is a screening one.
@Eduard_Korchmar Thanks for looking!

  • Yes, ICD10Pcs doesn’t allow us to differentiate the screening and the other mammographies. Also, counts for ICD10Pcs are too small to bother with applying additional rules to figure out if patients had screening.
  • Custom mappings can always exist, I don’t think there is much we can do except for checking such cases. 4147961 descendants may be a good catch, so we can simply limit it to the concept itself, without its descendants.
  • Yep, I think we want plain mammographies only.

(Michael Gurley) #9

This might be helpful:


(Christian Reich) #10

I think what @mgurley is trying to assert is that mammography by it’s nature is a screening procedure. You don’t use it to substantiate a breast cancer, you’d take a biopsy for that. I agree. Just collect all mammography concepts together.

@Eduard_Korchmar: Is your hierarchy going to do such a thing?

(Mark Danese) #11

Actually there are screening and diagnostic mammography codes in @mgurley’s link and in clinical practice. So, one would want to know whether just screening or both screening and diagnostic should be included. Which depends on the research question.

(Kristin Kostka, MPH) #12

Thanks @mgurley this is a very helpful resource. Based on Michael’s link, it looks like our concept set is mashing together mammography concept codes for both screening and diagnostic. So, let’s unpack and see what the right combination of codes are based on what we’re trying to do.

The study question is:

Amongst women aged 40-74 who undergo a screening mammography with no prior malignancies (Target Cohort 1) or amongst women aged 40-74 who undergo a screening mammography with prior malignancies (Target Cohort 2) and do not have prior breast cancer recorded or history of breast cancer, which patients will go on to develop breast cancer in the 90d to 3 years following the screening mammography?

Our clinical rationale:

Having two T risk groups will allow us to see how the presence or absence of prior malignancies affects the model. We want to be able to identify risk for two groups (healthy and those with prior malignancies) to see what factors influence getting a diagnosis of breast cancer. There are risk predictors of breast cancer currently and they are using age, prior birth age, first period, family history, biopsy and race. Our results could confirm that there are specific predictors related to what has been assessed, and confirm knowledge risk factors with data.

We’re focusing on the design principles of the concept set of how we’re building mammography. So according to Michael’s reference from SEER, we’re looking at:

Type of test Screening Codes Diagnostic Codes
Mammogram 76092 76090-unilateral, 76091-bilateral

Which in ATHENA brings us to:

Domain ID Procedure
Concept Class IDCPT4
Vocabulary ID CPT4
Concept ID 42737560
Concept code 76092
Invalid reason Valid
Standard concept Standard
Synonyms Screening mammography, bilateral (two view film study of each breast)
Valid start 12/31/1969
Valid end 07/02/2007

@Dymshyts @Eduard_Korchmar Not sure if the end date is an artifact or if it’s actually been depreciated. Is there a more appropriate code? Our intention is to ensure our Screening Mammography concept set is not riddled with misclassification bias. I don’t think we’ll ever get 100% here but let’s go for “good enough”… :slight_smile:

This is the last design debate before we can package and run the study! :smiley: We really appreciate the rapid community feedback. I heard a rumor you might see something presented at this year’s Symposium.

(cc: @aostropolets @MauraBeaton @rmakadia @ericaVoss @clairblacketer @Laura_Hester @sseager - can only tag 10 ppl, please loop in the other WoOs)

(Eduard Korchmar) #13

As long as concept is valid and standard, there should be no problems using it. It is kept standard on purpose since it is still present in source data and has no better mapping but itself.

However, since @aostropolets says you only need plain screening radiographies, perhaps codes for tomosynthesis should be excluded too? I’m talking about concept_ids 46257687,46257521,45890623.

Judging from info on this resource, it is a rather sophisticated procedure requiring specialized equipment: maybe it is unlikely this procedure is performed as screening measure?

(Christian Reich) #14

I am not tagged, so at the danger of not being wanted in the debate: :slight_smile:

The difference between screening and diagnostic is simple: Screening gets done on asymptomatic patients. Healthy women with no knots in the breast. The codes call out if it is for screening or not, but I am not sure this is reliable. But I wouldn’t know either way. Somebody needs to find out from CMS coding rules. In the rest of the world I am sure we cannot rely on that information. Also, there are a LOT MORE asymptomatic women with the procedure than with. So, we may have a few misclassifications, but I doubt they will have an effect on the study.

But at the end of the day the authors have to decide if they explicitly want asymptomatic women being screened, or all.

Hang on a second! You want to run two PLPs and then compare them? Compare the models for their “risk factors”? Won’t work, because they are regularized, and the absence of a covariate of one may not mean it has no predictive value. Or do you want to compare the risk of individuals to develop T1 vs T2? Not sure the statistical behavior of doing that on the whole population and then drawing inferences is kosher. You need @msuchard or @schuemie to weigh in on that.

(Eduard Korchmar) #15

None of the SNOMED concepts that specify radiography with screening intent are expected to have ICD10PCS descendants since imaging procedures in latter are not designed to specify intent.

Just checked: in current version no such links exist.
If we want a certain ICD-10-PCS procedure to be a descendant of radiographic screening in SNOMED, it has to be done manually and against our rules of not basing logic on assumptions.

(Anna Ostropolets) #16

@krfeeney what exactly are we mashing?
@Christian_Reich yes, we checked CMS coding rules. For example,

This concept set contains: billing codes for screening mammographies + SNOMED. Don’t want SNOMED - say it.

(Kristin Kostka, MPH) #17

Question answered: if a forum post caps out of people tags, will @Christian_Reich still see it and respond? :laughing:

Agreed. This study will likely struggle ex-US… but to be fair, we’re studying US practice guidelines so it’s a bit circular. We had debate over asymptomatic only and decided to look at all women in the desired age band.

Given our study question specifies “screening mammography”… I think we need to clean up the concept set to be more limited. @aostropolets, it looks like we’re mixing screening and diagnostic mammography procedure codes here.

For example, we’re using:

Is it appropriate to include these? Do we know something about practice patterns that suggests these are used for more than diagnostic purposes? Anecdotally we heard from a nice person who checked their own EMR that a patient with prior malignancy will still be given a “screening” mammography code not a diagnostic for a routine mammogram (as prescribed by the guidelines). So the question is… why do we include these in the concept set? Doesn’t seem to be part of our Target population, yeah?

As for the methods debate…I hear you, Christian. but I’m ignoring you until we actually get to having any results at all. I’m not choosy (yet). @jreps is our parental supervision here.

(Andrew Williams) #18

It is important though and a convention for representing screening intent at least in the US is a good idea. Any comparative effectiveness study between screening strategies has to be able to distinguish whether the test was done in the presence of symptoms or the study basically can’t be done in a valid way. You’ll have too many cancer or higher stage outcomes when the ratio of disease outcomes to tests is quite low. So those extra outcomes from non-screening uses have a strong impact on results. Being able to use codes without resorting to NLP to do ascertain intent will be very valuable when it can be done. In studies I’ve been involved with, coded intent was reliable for mammography but had to be done manually for screening modalities with the same ambiguity about intent like colonoscopy.

(Kristin Kostka, MPH) #19

Valid points, @Andrew, but not the course we’re charting. We’re not doing a comparative effectiveness of two cohorts. We’re looking at which patients will go on to develop breast cancer in the 90d to 3 years following a screening mammography. We think there is some differential classification between someone with prior malignancy versus not so we decided that merits two target cohorts to build two separate risk models.

(Anna Ostropolets) #20

@Andrew It’s a good point. But it seems like adding additional criteria will make the analysis way more complicated and time-consuming, right? Besides, CPT4 codes go together with the records with the results (like Patient information entered into a reminder system with a target due date for the next mammogram (RAD)), which, I’m assuming, favors the assumption that the mammography was a screening one.
@krfeeney here’s the explanation for G0206 and G0204. It used to be used as screening and then was replaced by CPT4 screening codes.Here is the explanation for G0279, which says that this code is used as add-on code (which also means that it was added to screening mammography).

I would love just to take one SNOMED code and its descendants. But it doesn’t work this way here, so we did some research on CMS guidelines and came up with the list (again, except for 2 SNOMED code).