Problem Statements

We started working on the TU/TR domains when this question was raised: if an abnormality can be considered as a lesion, should it be modeled using the TU/TR structure? This question came from modeling aneurysm data from the cardiovascular TA projects, where the aneurysm data were originally mapped to CV and discussions had taken place on whether it should be mapped to TU/TR. During our attempt to model the aneurysm and other related CV data to TU/TR, we noticed that: tumors are neoplastic lesions while an aneurysm can be considered as a non-neoplastic, cardiovascular lesion. They all are considered as lesions, why is the modeling for the two types of lesions so different and inconsistent in TU?

TU/TR/RS were originally created for tumor identification and (tumor) disease assessment data. Non-tumor lesion modeling was later added onto TU/TR. RS was also combined with CC in either 2018 or 2019.

Arguments have been made that Tumor finding data modeling is prefaced on “already identified tumors”. TU records are created for each “identified tumor”, with the creation of a unique tumor ID (TULNKID). Since only already identified tumors are mapped to TU, this means, the locations (produced by imaging procedures) where one looked for the presence of a tumor and didn’t find it, is NOT modeled in TU.

  • This is in part due to the fact that imaging location is treated as a procedure attribute and is mapped to procedure.

The results for TUTEST = Tumor Identification, are typically TUORRES = Target and Non-target. However results for this test can also be: BONE LESION, measurable/non-measurable, benign/malignant, new, etc. We think the purpose of this test is to identify the “role” of a tumor, such as tumor type, malignancy status, measurability, etc. However, TU domain definition also states that TU is used to identify the location of the tumor. There is no specific TUTEST associated with the tumor location identification process - tumor locations are treated as a known fact and are mapped to TULOC directly for the Tumor Identification TUTEST. It isn't clear whether TUTEST = Tumor Identification, is also intended to be used to identify the location of the tumor.

Identified tumors are considered to be target/non-target according to published criterion (e.g. RECIST) and this is to classify tumors for the subsequent assessment of disease response to treatment (RS).

This doesn’t work for non-tumor lesions, because:

  1. There is a need to model whether or not a non-tumor lesion is identified in a region.
    1. This requires imaging location to be modeled in the TU domain.
    2. TU modeling needs to allow the creation of a “negative” record – I looked at this location but didn’t find what I suspected.
  2. Lesion "Type" and "Location" identifications are a separate process and test from "Role" identification. During the location identification process, the location for an identified lesion is a result qualifier. When this lesion undergoes "role" assessment, the location of the identified lesion is then a test qualifier.
  3. Once the location for a non-tumor lesion is identified, this lesion may not be assigned a role (i.e. target) because said lesion will not undergo “assessment of disease response to therapy (i.e. RS)”. This lesion may be further classified based on severity, anatomy or other types of classifications according to a published, validated public domain or copyrighted criteria. In this case, this should be modeled in the CC portion of the RS/CC combined domain.

The concept map below describes the modeling discrepancies between tumor and non-tumor lesions. Note the dotted lines and boxes indicate a step that didn't happen and data not modeled.

TU Domain Modeling: Tumor vs Non-tumor Lesion Concept Map


Agreed Approaches so Far

  1. Imaging location should be treated as the TEST Location when identifying the location of a suspected object.
  2. We will not make changes to the tumor findings modeling in TU, several cancer TAUGs have been published and we do not want to cause backward compatibility issues.  We will only update the non-tumor lesions side.

Domain Level Questions

  • If all I need to do is to identify the type of a lesion, it’s location, characterize its physical properties (e.g. diameter measurement) and then evaluate its severity, there is no disease response to treatment data, in other words I have TU, maybe TR and CC data, but no RS, can I use the TU/TR structure?
  • How do we model Result Location, when the result itself is the location? Should this be modeled in variables (RESLOC), or should we create tests for result location findings?
  • Lesion "Type" and "Location" identification should be modeled separately from "Role" Identification.
  • Just because something can be considered as a lesion, should it be modeled in TU/TR? 

Recommendations

  1. Combining domains cause modeling inconsistencies. Oftentimes domains are created by different groups under different circumstances. The premises based on which a domain is built varies greatly for essentially, the same type of data (Tumor vs Non-tumor lesions). Because domains are governed by different teams (i.e. oncology and QRS SDS/CT teams), there are also different CT business rules developed for different domains.
  2. May consider adding an assumption to better define "lesion" for TU: A localized pathological or traumatic structural change, damage, deformity, or discontinuity of tissue, organ, or body part (C3824). Anything that causes fundamental architectural changes to the structure is considered as a lesion. Jordan Li : ask Al and Nick to draft a better definition for lesion.
  3. Consider updating the TUTEST = Tumor Identification to: Tumor "Role" Identification




  • No labels

2 Comments

  1. Jordan Li

    Currently not in scope for LOC team but has been raised lately. Discussion with the LOC subteam needed.

    What should go into PRLOC?

    Email to Erin:

    I was looking through the procedure codelist to see what diagnostic imaging procedure terms we have published and noted something interesting. For example, we have:

    AORTIC VALVE REPLACEMENT
    MITRAL VALVE REPLACEMENT
    PULMONARY VALVE REPLACEMENT
    TRICUSPID VALVE REPLACEMENT

    Their definitions are nearly identical (swap out the affected valves), I was wondering why PRLOC cannot be used to map the four cardiac valves? Why do these terms all have their locations pre-coordinated?


    Erin:

    I think we need to think about the granularity of what the procedure codelist should look like w.r.t. the other variables present in the domain.

    Something like ‘Aortic Valve Replacement’ has an individual CPT code so if you were using CPT as your coding system for PRDECOD, you would pre-coordinate that information into the DECOD value and not necessarily use other variables in the domain. If the CDISC codelist chose to only use more generic types of procedures, then one may need to use those additional variables. This choice of which coding system to use could end up complicating the PR dataset though presumably one would choose a single coding system to use.

    I think this requires some discussion within the LOC subteam to determine if there is a more logical way. We could always retire those existing concepts, create a single ‘valve replacement’ concept, and then put LOC information into the LOC variable. The team would need to talk that through though and consider how sponsors are currently doing this.

  2. Diane Wold

    It helps me to think of a findings record as representing information about both an observation (test) and observation result (result).  The test location is an attribute of the observation, the result location is an attribute of the observation result.

    The oncology test "tumor identification" is actually a complex activity that starts with an observation which locates tumors.  If that were all, the test would be similar to one of the ECG tests for which results are the names of abnormalities.  I'm not sure whether the result values would all be "TUMOR" or something more specific.  The "tumor identification" test goes also includes two more items, the assignment of an evaluator-specific tumor identifier (in TULNKID) and the assignment of a role for response assessment (in TUORRES).  The first stage result is actually incorporated into the test name.  There are similar tests of "Lesion Identification", "Limb Lesion Identification" and "Vessel Lesion Identification".

    Some other TU tests are more straight forward.  A number are indicators that ask whether a particular kind of abnormality was found.  This is similar to indicator tests in other domains.

    The Disease Recurrence Relative Location test is used in cases where it is already known that a subject has a disease recurrence, and asks whether the recurrence is local, regional, or contralateral. (I think those are the responses, but my memory may be wrong.)

    The other TU test that deals with location is Primary Site Indicator, which I think was a mistake.  This is really about whether which tumor is the primary tumor, but can beused in cases where the primary tumor was resected before the study. I would prefer representing this as a attribute of or test about the primary tumor, even if this means creating a pre-study record for a primary tumor about which little data is collected for the study.  Is whether a tumor is primary a TR test? What about a test which asks whether the tumor was resected?  Is that an occurrence test about a procedure?