Amendment History

Table of Contents
1 Glossary
1.1 Domain Terms
2 Introduction
2.1 Purpose
2.2 Audience and stakeholder domain
2.2.1 Input from stakeholders
2.2.2 Degree of consensus on the statement of problem
3 Statement of the problem or need
3.1 Summary of problem or need, as reported
3.2 Summary of requested solution
3.3 Statement of problem as understood
3.4 Detailed analysis of reported problem, including background
3.4.1 18-p syndrome with associated immunodeficiency (disorder)
3.5 Subsidiary and interrelated problems
4 Risks / Benefits
4.1.1 Risks of not addressing the problem
4.1.2 Risks of addressing the problem
5 Requirements: criteria for success and completion
5.1 Criteria for success/completion
5.2 Strategic and/or specific operational use cases
5.2.1 Use case 1
5.2.2 Use case 2
6 Solution Development
6.1 Initial Design
6.1.1 Outline of initial design
6.1.2 Significant design or implementation decisions / compromises
6.1.3 Evaluation of Design
6.2 Iteration One
6.2.1 Outline of design
6.2.2 Evaluation of Design
6.3 Iteration Two
6.3.1 Outline of revised design
6.3.2 Significant design or implementation changes
6.3.3 Evaluation of Revised Design
6.4 Iteration Three
6.4.1 Outline of revised design
6.4.2 Significant design or implementation changes
6.4.3 Evaluation of Revised Design
7 Recommendation
7.1.1 Detailed design final specification
7.1.2 Iteration plan
8 Quality program criteria
8.1 Quality metrics
8.1.1 Quality metric 1
8.1.2 Quality metric 2
9 Project Resource Estimates
9.1 Scope of construction phase
9.2 Projection of remaining overall project resource requirements
9.2.1 Expected project resource requirement category
9.2.2 Expected project impact and benefit
9.2.3 Indicative resource estimates for construction, transition and maintenance:

Glossary

Domain Terms

Introduction

Purpose

The purpose of this project is to consider that the attribute HAS DEFINITIONAL MANIFESTATION requires review for reproducibility and consistency. In October 2009, this attribute is used 288 times in the stated view but impacts 4600 concepts in the inferred view.
SNOMED CT projects transition from Inception Phase ? Elaboration Phase ? Construction Phase ? Transition Phase. This document combines the documentation of the Inception and Elaboration Phases.
The Inception Phase focuses on understanding the problem and its scope, identifying stakeholders and their requirements, and identifying risks.
The purpose of the Elaboration Phase is to develop, document and test one (or more) possible technical solutions, and to reach a recommendation and provide a detailed specification of a preferred solution to be taken forward to the construction phase.

Audience and stakeholder domain

The audience for this document includes all standards terminology leaders, implementers and users but is especially targeted at those stakeholders with an interest in doing quality control on disorder attributes in SNOMED CT.
A further significant audience is the community of SNOMED authors that may be requested to implement the recommended specification.
This request was submitted to the Consultant Terminologist Program by Kent Spackman in March of 2010.

Input from stakeholders

The HAS DEFINITIONAL MANIFESTATION attribute is not being used in an understandable, reproducible, and useful fashion.

Degree of consensus on the statement of problem

The statement of the problem is clear, and my conversations with modelers indicate a shared concern regarding lack of reproducibility of this attribute, there is still sense among some that there is utility in the attribute if we can improve how it is used. The editorial guide says that [t]his attribute links disorders to the manifestations (observations) that define them. It can only be applied to disorders.
Unfortunately, a reproducible distinction upon what constitutes a disorder has been an elusive challenge, and it is also challenging to know when an observation is merely obvious and does not need to be defined, and when it should be defined in an URU fashion. For example, Peutz-Jeghers syndrome (disorder), is a disorder in SNOMED, which is caused by mutations in the STK11 (also called LKB1) tumor suppressor gene located on chromosome 19p13. Thus an observation that the patient had this mutation manifest would be expected, but there is no HAS DEFINITIONAL MANIFESTATION relationship associated with Peutz-Jeghers syndrome (disorder) within SNOMED at this time, and there is no concept associated with STK11 gene mutations, but there are concepts associated with other genetic mutations such as BRCA1 and BRCA2.
Should new observations, such as observations of STK11 mutation or BRCA1 mutation be added so that HAS DEFINITIONAL MANIFESTATION can be reproducibly applied?

Statement of the problem or need

Summary of problem or need, as reported

The following description is provided on the IHTSDO issue tracker:
The attribute HAS DEFINITIONAL MANIFESTATION requires review for reproducibility and consistency. In October 2009, this attribute is used 288 times in the stated view but impacts 4600 concepts in the inferred view.
The SNOMED Editor guide defines HAS DEFINITIONAL MANIFESTATION as an "attribute [that] links disorders to the manifestations (observations) that define them. It can only be applied to disorders…" Note that a definitional manifestation is not necessarily pathognomonic If it were explicitly stated that it meant pathognomonic findings, would that change your mind about it?If it where defined as being pathognomonic, then I would change my interpretation, however most of the definitions I've seen may be tautological, but are not pathognomonic. So the hierarchy would still have to be reworked, with dubious value resulting from that work.
Tautological data just adds complexity without improving understanding. (pathognomonic meaning that a clinical sign or lesion is so characteristic of a specific disease that one can make a diagnosis—definitive or highly suggestive diagnosis—based on simply seeing the sign or lesion). I make this clarification because I initially mistakenly thought the definitional manifestations where intended to be pathognomonic.

Summary of requested solution

There is no specified solution other than to review use of Has Definitional Manifestation, to determine if it is understandable, reproducible, and useful, or to make recommendations concerning the disposition of HAS DEFINITIONAL MANIFESTATION. My review of the SIRS tracker has 7 related issues:Please could we have the full SIRS item referenced (maybe a table with item number, request description and reason – could be an appendixDone

1. Fredrickson lipidemias: 393998
2. Disorder - def manifestation: 392442
3. rubella deafness: 348108
4. Functional abdominal pain syndrome (disorder): 348107
5. Disorder characterized by back pain (disorder): 348106
6. Painful blind eye (disorder): 348105
7. Osteoarthritis of multiple joints (disorder): 6034

My review of these trackers are they wish to provide a more specific definition, and propose use of HAS DEFINITIONAL MANIFESTATION, but they don't specifically give reasons why use of this attribute is a better solution that others that we might consider. Some also state that use of HAS DEFINITIONAL MANIFESTATION is on hold until the reproducibility of the attribute can be sorted out. I believe a more reproducible and useful solution is to use is-a, in most cases, instead of HAS DEFINITIONAL MANIFESTATION, or to eliminate the relationship in others, as I will describe in this document.

Statement of problem as understood

From this problem statement, it is inferred that the attribute Has Definitional Manifestation has likely been overused or incorrectly used, leading one to believe that the basic operational criteria of URUmay no longer apply to Has Definitional Manifestation. Concepts with this attribute will need to be analyzed for appropriate use, and likely have attributes modified to regain alliance with this overarching principal for SNOMED CT content creation.

Detailed analysis of reported problem, including background

A search query was performed, showing that 3063 concepts in SNOMED CT include a relationship type of Has Definitional Manifestation. Subsequent metrics have shown some clean up of usage of this attribute, including reduction in the number of stated circular relationship (a relationship pointing to the parent of the concept it defines) as shown in Figure 1. *Error! Reference source not found.*Something missing hereShould link to figure 1.

Figure 1. Circular stated relationship counts over time for "Has definitional manifestation"

From this set, some concepts where selected to examine if the use of the Has Definitional Manifestation is URU. Consider for example, Adrenocorticotropic hormone hypersecretion (disorder) has a definitional manifestation of Increased hormone production (finding).

Figure 2. Inferred definition of Adrenocorticotropic hormone hypersecretion (disorder).
Increased hormone production (finding) has ~45 concepts that use it for the restriction of has definitional manifestation relationships. It's sibling, Decreased hormone production (finding) has ~15 concepts that use it as a restriction of the has definitional manifestation relationships. Notably absent from this list is Diabetes mellitus type 1 (disorder) which is simply defined as:

Figure 3. Inferred definition of Diabetes mellitus type 1 (disorder).
Where I believe that consistent use of the Has definitional manifestation would require such a relationship to Decreased hormone production (finding) Or 131104004 Decreased insulin level (finding)Well, decreased hormone level is different than decreased hormone production… Level is a manifestation regardless of cause. Perhaps the level is low because an insulin blocker artificially increases the potency of insulin, meaning that you need secrete less to prevent hypoglycemia. For example: http://www.nature.com/news/blocking-insulin-breakdown-shows-promise-against-diabetes-1.15273for Diabetes mellitus type 1 (disorder).
Based on current usage, application of the Has definitional manifestation attribute is at least incomplete (not used everywhere it can be used), if not inconsistent (the patterns of usage are applied more rigorously in different areas of the taxonomy than in others). I think that definitional manifestation is so broad that essentially every observation (not just diseases) should have the relationship if it were to be retained. Ways to improve consistency would then require automated support through QA rules requiring specification of this relationship, or requiring that the absence of the relationship be white listed. But again, this is such a large task, and would require many new concepts (such as findings of genetic mutations as I discussed aboveThis issue has been discussed by ECE – Bruce is joining us and can elaborateYes, it was a good discussion, but retiring the Has definitional manifestation relationship secondary to inconsistency and incompleteness should be neutral with regard to the ECE model. ), I think we are better off splitting the problem into smaller more manageable chunks, and reconsider how Has definitional manifestation contributes to the overall structure, and consider if the cost of maintaining this relationship proportional to the benefit of doing so. Based on my analysis and experience, there is a better way to go forward, by reconsidering the use and interpretation of is-a in the context of findings and diseases.
Given current and historical challenges making reproducible distinctions between clinical findings, observations, problems and diseases (disease currently is-a clinical finding whereas there was a historical effort to make disease and clinical finding disjoint) there is an uncomfortable recursion of sorts, where the terminologist might have to debate if an is-a relationship or a has-definitional-manifestation relationship—or both—might be appropriate for relating two clinical finding concepts to each other. An example of the challenges would be that you are modeling a finding (not a disease/disorder), but you believe that the concept you are modeling meets the criterion for a disease/disorder? Having to make judgments about disease/disorder just complicates the task, as you cannot (currently) apply definitional manifestation to a finding that is not a disorder.

18-p syndrome with associated immunodeficiency (disorder)

18-p syndrome with associated immunodeficiency (disorder) is a representative example Is this representative of the majority of existing content with 'has definitional manifestation or a relatively special case? We should use simple and concentrate on the common but with some special cases. So I think I am arguing for keeping this but adding a few simple examples first.This is a representative example of the majority of the content. More complete examples are included in the accompanying document, Stated usage of "has definitional manifestation," recommendations for update, and identification of concurrent issues that demonstrates the challenge associated with modeling Has definitional manifestation.
In many cases, the use of the Has definitional manifestation attribute is very general, for example saying that 18-p syndrome with associated immunodeficiency (disorder) Has definitional manifestation
Immune system finding. Figure 4. Original stated definition of 18-p syndrome with associated immunodeficiency (disorder)(below) shows the stated definition of 18-p syndrome with associated immunodeficiency (disorder) and Figure 5 (below) shows the inferred form of this concept. Note that the Has definitional manifestation relationship to Immune system finding (finding) is an inherited relationship. But why not model the concept by saying that 18-p syndrome with associated immunodeficiency (disorder) Is a Immune system finding (finding)?


Figure 4. Original stated definition of 18-p syndrome with associated immunodeficiency (disorder)

Figure 5. Inferred form of 18-p syndrome with associated immunodeficiency (disorder) prior to adding an is-a relationship to Deletion of short arm of chromosome 18 (disorder). Note that the Has definitional manifestation relationship is inherited from an ancestor concept, specifically Immune system finding (finding).
I think we would be better off stating that 18-p syndrome with associated immunodeficiency (disorder) Is a Immune system finding, rather than stating that it Has definitional manifestation Immune system finding. This use of the Is a will simplify class-based retrieval of the concepts, and reduce retrieval errors. For example, under current definitions, querying for an Immune system finding will fail to retrieve 18-p syndrome with associated immunodeficiency which most users would consider a Type II error (a failure to detect something that is present).
I'll further illustrate this recommendation, by considering how 18-p syndrome with associated immunodeficiency (disorder) should relate to the concept Deletion of short arm of chromosome 18 (disorder). The two options are to use an Is a relationship, or to use a Has definitional manifestation.
Lets consider the case of Is a first. Ffollowing this is-a modeling style, the definition of 18-p syndrome with associated immunodeficiency (disorder) can be modeled as shown below in Figure 6.

Figure 6. Proposed changes to stated form of 18-p syndrome with associated immunodeficiency (disorder)
Committing and classifying this change results in a much richer inheritance, that will be the case for adding the Has definitional manifestation relationship, as we will show next in Figure 8.


Figure 7. Inferred form of 18-p syndrome with associated immunodeficiency (disorder) after adding an is-a relationship to Deletion of short arm of chromosome 18 (disorder).

Figure 8. Adding a relationship to Deletion of short arm of chromosome 18 (disorder) as a Has definitional manifestation relationship instead of as an is-a relationship.

Figure 9. Inferred form of 18-p syndrome with associated immunodeficiency (disorder) after adding a Has definitional manifestation relationship to Deletion of short arm of chromosome 18 (disorder).
Compare the resulting inferred formOne of the key elements here is the determination of causality. The ECE group has come up with naming patterns to address this. If, as is probably the case here the immunodeficiency is due to the genetic anomaly, then this would be "Immunodeficiency co-occurrent and due to deletion of short arm of chromosome 18". Te modeling would then have both a DUE TO and IS-A relationship to "Deletion of short arm of chromosome 18"Yes. As discussed by Monique and myself on previous calls… I'm uncomfortable with "due-to" within the T-Box, but that's a tangential issue. I agree according to the ECE group you would end up with this relationship, so I'm not wanting to challenge that assumption in this document. Because DUE TO causes both IS-A and DUE-TO relationships within the same taxonomy, and cannot be kept disjoint, I think you'll end up with a never ending cascade potential tautolicical and circular relationships.
I really don't want to conflate "has definitional manifestation" and "due to"… They are both interesting issues, but I want to keep this discussion wtihin the scope of "has definitional manifestation.", shown above in Figure 9, with the classification from adding an Is a relationship to Deletion of short arm of chromosome 18 (disorder) shown in Figure 7. Note that the inheritance is more complete when an Is a relationship is used than it is when a Has definitional manifestation relationship is used. The more complete inheritance that results from using an Is a relationship will result in less Type II retrieval errors.
This simple substitution of is-a for Has definitional manifestation will require reconsidering some of the finer points of how is-a is interpreted. Consider the concept Epilepsy (disorder), which has definitional manifestation SeizureThis is another pattern that relates to the proposed SDP project, where a person with epilespy has a predisposition to seizures. The name of the disorder in this case is not a seizure, but a moniker to describe recurrent seizure events.I like the general notion of predisposition. Having effective guidelines to that such relationships don't become tautolicical and circular relationships is important. Should "senile keratosis" be "due-to" advanced age? And sun exposure? And ultravoilet radiation? Or non-compliance with sunscreen recommendations? Or failing to wear long-sleeved shirts and a wide brimeed hat?
Should you model that "sun exposure" predisposes to malignant melanoma, sunburn, and that sun exposure increases hormone production for Vit D?
I believe these are all true and useful, but I believe they belong in the A-Box, not in the SNOMED T-Box…:

Figure 10. SNOMED representation of Epilepsy (disorder).
Traditionally, Kent has taught us that is not true that a Seizure disorder is a Seizure any more than it is true that a single measurement of a systolic blood pressure of 140 means that the patient has a hypertensive disorder. My review of the Has definitional manifestation uses shows that there are many uses that may be better represented as is-a relationships Yes, agree that many are better off as ISA. But your statement below indicates that this shouldbe evaluated on a case by case basis.Yes, which I did in the accompanying ~400 page document, where I reviewed each definition, and made specific recommendations for which ones should be is-a, which ones should be retired, etc. (central cyanosis is-a cyanosis), and some that would be improper to represent as is-a relationships given the traditional teaching (seizure disorder is-NOT-a seizure). Perhaps it is time to reconsider this specific teaching about what is-a means within the Clinical Finding taxonomy and elsewhere. After evaluating several thousand of the Is a and has definitional manifestation attributes, I came up with the following thought process that I believe simplifies, and makes more reproducible the content development and usage process with no reduction in the benefit that the description logic provides.
Is-a requires that everything that is true of the ancestor concept is true of the concept being defined. Consider that the finding taxonomy refers to findings—not EVENTS. When considering the relationship between Epilepsy (disorder) and Seizure (finding), consider if the manifestations of having a finding of an Epilepsy disorder include all of the manifestations of a seizure finding… I propose that the answer is true… In order to have an Epilepsy disorder, the patient must have had all the manifestations of a Seizure (finding), within the time span that the Epilepsy disorder is determined to be present. A seizure does not require that an Epilepsy disorder be present, but an epilepsy disorder does require that at least one instance of a seizure finding be present, so I propose that it is true that a "finding of a seizure disorder is-a finding of a seizure"… Note that this definition does not imply a specific count of seizures other than at least 1… But there had to be at least 1 finding of a seizure in order for a patient to have a seizure disorder. The is-a relationship is not a quantitative relationship, other than implying that all manifestations of the ancestor be present in the concept being defined.
Given this interpretation, how do we capture that with a hypertensive disorder, high blood pressure is observed,This is an example where the INTERPRETS/HAS INTERPRETATION seems to fit. The differentiating feature is that this is a disorder, so it would classify as one as opposed to only a finding.Yes. Someday it will be nice to consider a grand-unified theory where observation/finding/disorder pull together in a more satisfying way… But I guess Physics has been trying to find unified theories, and it has taken some time. So I guess I need patience, while we work through our evolution of thoughts about modeling. ?? We capture that with and is-a to high blood pressure finding: A finding of a hypertensive disorder is a finding of elevated blood pressure, although a specific event of a particular blood pressure measurement is not implied. It is only implied that the finding of elevated blood pressure was necessarily present during the episode of the hypertensive disorder.
I presented this idea of a broader interpretation of Is a to the consultant terminologist group, who where generally unimpressedI concur with the consultant terminologists suggestion that a wider interpretation of 'IS_A' should be avoided but nothing to do with being unimpressed – I just think we should be as precise as possible in our definitions and use and if a wider definition is required, should we consider adoption of another relationship in addition to IS_A.I'm using unimpressed as is described by dictionary.com as: not having a favorable opinion: unimpressed by his arguments. This is a perfectly OK outcome. I'm personally concerned that use of a different relationship in addition to IS_A would be counterproductive with regard to representational complexity, so I'm just noting this here, so the thought process is documented, and it may be useful for others to think about the same issues. . So for the purposes of this elaboration phase, the broader interpretation will not prevail, but the basic idea of turning most of these Has definitional manifestation relationships into Is a relationships was accepted. I choose to leave the broader discussion of the use of Is a in this document as a foundation for future discussions on the topic.

Subsidiary and interrelated problems

Some ongoing challenges in creating consistency in the approach to definitional manifestation include:

• Challenges reproducibly defining the differences between observations, findings, problems, diseases, and/or disorders. The proposed approach simplifies simplifies the inherent challenges by not requiring a distinction between observations, finding, problems, and/or disorders. The proposed approach just requires that you consider if the observations, finding, problems, and/or disorders have all the manifestations of the ancestor concept at least once during the interval during which the observations, finding, problems, and/or disorders are considered present.
• Allowing defining relationships other than Is a within a discrete taxonomy (when do you use Is a vs Has definitional manifestation?). However, the challenges associated with trying to make observations, findings, problems, diseases, and/or disorders reproducibly discrete have so far been difficult. The proposed approach simplifies the problems with potential circular relationships by eliminating one of the relationships that have the potential for circular definitions, simplifying the modeling task, as well as the use/implementations tasks.

Risks / Benefits

Risks of not addressing the problem

Currently Has definitional manifestation is at least incompletely applied, if not inconsistently applied, and in some cases is applied when an is-a relationship might be more appropriate. This creates challenges when performing class-based retrieval of data, and particularly results in Type II retrieval errors.

Risks of addressing the problem

Change makes some people sorry, sad things (to quote Dar Williams). There will be inevitable disruption to the way people conceptualize and use SNOMED, good and bad.

Requirements: criteria for success and completion

Criteria for success/completion

A currently large (if not unbounded) modeling challenge of consistently applying Has definitional manifestation across the entire finding taxonomy will be reduced to a number of smaller modeling activities that can be limited in scope, while still contributing to the correctness and completeness of data retrieval that uses the SNOMED description logic for concept retrieval. [Note: Another criteria for success would be the lack of incorrect/overstated inferences where IS-A relationships replaced HAS DEFINITIONAL MANIFESTATION.]

Strategic and/or specific operational use cases

The description logic representation of SNOMED needs to be understandable, reproducible, and useful. This project will incrementally improve the understandability, reproducibility, and utility of SNOMED. This improvement will make it easier to use SNOMED CT, and easier to model SNOMED CT. Specific use cases include any kind of algorithmic decision support that depends on the SNOMED semantic definitions for class-based query, or other kinds of semantic retrieval.

Use case 1

Class based queries will properly return all concepts in that class. This change will reduce the number of Type II errors associated with class based queries.

Fit with IHTSDO strategy

Class based queries are a cornerstone of the intended uses of SNOMED for decision support, or predictive analytics based on data encoded using SNOMED.

Use case 2

The definition of concepts is understandable, reproducible, and useful. This change will result in simpler modeling that will therefore be more reproducible.

Fit with IHTSDO strategy

Coherent logical definitions, modeled in an understandable, reproducible, and useful way is essential to support intended uses of SNOMED.

Solution Development

Initial Design

Outline of initial design At the end of this example used to demonstrate the design suggests that it is ambiguous and should perhaps be retired. If this is the case the example doesn't demonstrate the designThe heading is from a template, and I think better words than design might be used, but I chose to accept it. In this context it is the design of the approach to determine if the Has definitional manifestation relationships are ambiguous and should be retired.

It is a little different than designing an approach to improve the taxonomic classification of acute inflammatory disease, but still fits reasonably within the range of interpretation of design.
Review each of the Has definitional manifestation relationships, and consider which ones might be converted to is-a relationships. For the remaining Has definitional manifestation relationships, determine if they might be able to be replaced by improvements/additions in other areas of SNOMED. For example, for Angle-closure glaucoma (disorder): This is an interesting case because the HAS DEFINITIONAL MANIFESTATION adds nothing to the inferences. In fact, this specialization of glaucoma is sufficiently defined by just the primitive parent and the relationship group. So in this case, the HAS DEFINITIONAL MANIFESTATION relationship can be retired.Yes, a tautological relationship…

Figure 11.
If the morphology concept Obstruction (morphologic abnormality) had children such as Angle closure obstruction (morphologic abnormality) would the need to represent Has definitional manifestation still exist? Further, on examining this particular finding, I noticed that there were two different relationships pointing to angle closure from the same concept Primary angle-closure glaucoma (disorder):In this case, the addition of "primary" is not really supported, but we are looking at extending the range of the PATHOLOGICAL PROCESS attribute to include "Idiopathic" which could then differentiate this from its parent. Again, here the HAS DEFINITIONAL MANIFESTATION relationship adds nothing to the inferences or the meaning.Yes ?

Figure 12.
One might ask if this a modeling error with the selection of the wrong value for definition manifestation? The closed angle is the cause not the manifestation. The closed angle is directly observable via Gonioscopy and potentially Biomicroscopy (slit lamp), so it is a manifestation that is directly correlated to a cause…
Which just makes assumed meaning of Angle closed (finding) In this case, Angle closed is an observation (i.e. finding) that may or may not have additional pathological effects. In remodeling this with an ASSOCIATED MORPHOLOGY of obstruction and making it fully defined, we see:
which is a much better inference IMHO.Yes, I agree ?more challenging to understand. Perhaps this concept is ambiguous and should be retired and replaced (or not replaced).
To make the process tractable, the Has definitional manifestation relationships need to be simply reviewed for the appropriates for conversion to an Is a relationship, and if appropriate, converted, and if not, deleted.

Significant design or implementation decisions / compromises

Evaluation of Design

This design was presented to the consultant terminologist group, which evaluated the positive and negative aspects of the problem.

Exceptions and Problems

The criterion for determining if an Is a relationship is an appropriate relationship was appropriate was felt to be overly aggressive.Do we have a full listing of these criteria anywhere?Unfortunately no, outside of a possible recording of the session. For example, criterion for accepting that a finding of seizure disorder is a finding of seizure was felt to be too aggressive.

Design Strengths

SNOMED will be more URU in its representation, and it is easier to implement.

Design Weakness

No design weaknessesThe primary design weakness is a lack of solid criteria for when the relationship should be converted or retired or replaced by other relationships. Some guidance in this area would need to be developed before wholesale application. Testing of the proposal on a random sample of existing content would assist in developing those guidelines.Well, I think the criterion for retiring the relationship is pretty clear: always retire the has-definitional-manifestation relationships. I agree there are some uncertainties with regard to modeling is-a, due-to, interprets, and many of the SNOMED approved attributes. The review of correct modeling "as a whole" while retiring these relationships represents an opportunity for improving the definition as a whole.
However, please don't hold me accountable to give clear criterion for when is-a, due-to, interprets, and many of the SNOMED approved attributes other than has-definitional-manifestation. I don't believe such clear criterion exist. , just a modification of the criterion for replacing a Has definitional manifestation relationship with an Is a relationship was requested.

Design Risks

Iteration One

Iteration one consisted of reviewing all the stated Has definitional manifestation relationships, and determining which once should be replaced with Is a relationships. On the initial iteration, it was determined that all relationships except those relationships on the concepts:

1. Bisexual
2. Female homosexual
3. Male homosexual

are to be converted to Is a relationships.

Outline of design

The design of iteration one is from the original plan.

Evaluation of Design

The design was evaluated by review with the consultant terminologists group as part of a web conference.

Exceptions and Problems

Design Strengths

Simple to implement

Design Weakness

Overly aggressive converting Has See my previous comment on proper disposition.Please see my previous comments on not expanding scope beyond proper disposition of the has-definitional-manifestation relationships. They should always be retired. definitional manifestation relationships to Is a relationships.

Design Risks

Iteration TwoIt would be helpful if you could demonstrate the solution on a small section of the hierarchy which has a number of concepts which would be changed as a result of this proposal and then run it through the classifier to see what impact it has.

The accompanying document, Stated usage of "has definitional manifestation," recommendations for update, and identification of concurrent issues, provides much more details on magnitude of the issue including stated and inferred forms of all concepts directly affected by removing Has definitional manifestation relationships.
Iteration two was performed after feedback from the consultant terminologist group.

Outline of revised design

Significant design or implementation changes

The original plan is modified to not use Is a relationships between concepts such as Seizure disorder and Seizure.If the criterion has changed (we assume they have) the updated criteria should be listed here.Added more specific criterion. So the fact that a Seizure disorder must have had all the characteristics of at least one seizure event is not sufficient to use as an Is-a relationship, where the definition of Is-a means that all features of the parent must be present in the child.

Evaluation of Revised Design

In the revised design, it was determined that all relationships except those relationships on the concepts:Is it possible to group these into reasons why they have been excluded?Added a reason at the bottom. There is also further notes in the "has definitional manifestation," recommendations for update, and identification of concurrent issues, document.

1. Bisexual
2. Female homosexual
3. Male homosexual
4. Seizure disorder
5. Angle closure glaucoma
6. Anxiety disorder
7. Hematuria syndrome
8. Gout
9. Irritable bowel syndrome characterized by constipation
10. Irritable bowel syndrome variant of childhood with constipation
11. Fetishistic transvestism
12. Perinatal cyanotic attacks
13. Chronic steatorrhea
14. Hypertensive disorder, systemic arteria
15. Chronic hypercapnic respiratory failure
16. Hypertensive lower esophageal sphincter
17. Cholestatic jaundice syndrome
18. Heavy chain disease
19. Alpha heavy chain disease
20. Monoclonal gammopathy
21. Kappa light chain myeloma
22. Lambda light chain myeloma
23. Light chain monoclonal gammopathy of uncertain significance
24. Light chain myeloma
25. Gamma heavy chain disease
26. Immunoglobulin A monoclonal gammopathy of uncertain significance
27. Immunoglobulin A myeloma
28. Immunoglobulin D monoclonal gammopathy of uncertain significance
29. Immunoglobulin D myeloma
30. Immunoglobulin G monoclonal gammopathy of uncertain significance
31. Immunoglobulin G myeloma
32. Immunoglobulin M monoclonal gammopathy of uncertain significance
33. Mu heavy chain disease
34. Neutropenic disorder
35. Panic disorder
36. Phobic disorder
37. Periodic limb movement disorder
38. Autoimmune neonatal thrombocytopenia
39. Neonatal thrombocytopenia
40. Thrombocytopenic disorder
41. Protein-losing nephropathy
42. Tachyarrhythmia
43. Tic disorder
44. Vertiginous syndrome
45. Visual epilepsy
46. Masquerade syndrome
47. Neoplastic masquerade syndrome
48. Sun-induced wrinkles

are to be converted to Is a relationships. These concepts are not recommended to use is-a because in my editorial opinion, and given current guidance, I do not believe that the origin concept of the Has definitional manifestation has all features of destination relationship.

Exceptions and Problems

None noted.

Design Strengths

SNOMED will be more URU in its representation, and it is easer to implement.

Design Weakness

None noted.

Design Risks

Possibility of incorrect Is a assignment or errant editing as is the case with any editing task.

Iteration Three

Outline of revised design

Design is the same as previous, but a reproducibility review has been completed to determine reproducibility of the design.

Significant design or implementation changes

None.

Evaluation of Revised Design

As part of Iteration Three, there where 324 concepts that where reviewed, which had Definitional Manifestation as part of it's stated definition. 
Previously, all 324 concepts, and made recommendations for how to represent each of those concepts, and created a master document that included the stated form, inferred form, and recommendations for each concept, along with table of contents for with concepts to merge, which concepts to convert the has-definitional-manifestation to an is-a relationship, and concepts with other issues. 
To test reproducibility, we used three experienced modelers, two of which have both experience modeling SNOMED as their primary job for more than several years, and who have gone through the IHTSDO consultant terminologist project (Monique and Sarita), and 1 individual who has several experience modeling SNOMED as their primary job, has continued being responsible for managing terminology content (responsible for MeSH and RxNorm, with other related activities at NLM, but has not gone through the consultant terminologist project (John Kilbourne). 
John Kilbourne completed a complete review of the 324 concepts. He agreed with my recommendations 68% of the time. 
Both Monique and Sarita reviewed a significant number of concepts, more than sufficient for a statistically significant sample, but secondary to time limitations, they did not review every concept. To incorporate their work, I created an algorithm that determined the percent agreement over all opinions where either Monique or Sarita, or both rendered an opinion in addition to the opinion that John and I had previously rendered. Each concept was scored based on the percent agreement. 100% if all agreed with the original opinion, 75% if 4 opinions where rendered and 3 out of 4 agreed with the original premise, 50% if 4 opinions where rendered and 2 out of 4 agreed with the original premise, and 25% if only one agreed with the original premise. In situations where only Sarita or Monique rendered an opinion in addition to John and I, if 2 out of 3 agreed with the original premise, the score was given a 66%, and if only 1 agreed with the original premise, a score of 33% was given. The end statistic is then a weighted average with the percent agreement with the original premise. 
The average score for agreement among all modelers who rendered an opinion was: 78% of the time.  So on average, given a larger group of opinions, they were more likely to agree with the original recommendation, than was the case with simply agreement between John and myself on all concepts reviewed. 
Next, I considered how likely they were to agree amongst themselves, when 2 or 3 of them rendered opinions. There were several possible states as follows: 
YY:   100% agreement with each other, 100% agreement with the original recommendation
NN:   100% agreement with each other, 0% agreement with the original recommendation
YN:   0% agreement with each other; 50% agreement with the original recommendation 
YYY: 100 % agreement with each other; 100% agreement with the original recommendation
YYN: 66 % agreement with each other; 66% agreement with the original recommendation 
YNN: 66 % agreement with each other; 33% agreement with the original recommendation 
NNN: 100 % agreement with each other; 0% agreement with the original recommendation
The average score for agreement with each other is 77%; So to summarize, as a group of 4, we agree with the original recommendations 78% of the time, and as independent reviewers, they agreed with each other 77% of the time.  
My interpretation of this data, is that an agreement much higher than 78% is not really possible given that they can only agree with each other 77% of the time. If on the other hand, they were able to agree with each other 90% of the time or higher, it would mean that they were unified in their disagreement with the original recommendations, and that we should consider modifying the recommendations to be more reproducible. Such unity was not the case, and I believe it reflects the inherent ambiguity of the original relationship "has definitional manifestation" and its previous implementation/use. 

In addition, I think this data fully supports elimination of the has-definitional-manifestation relationship, and gives an option for implementation, such that where there was 100% agreement, the recommendation for the particular concept should be accepted, and in areas where there is lesser agreement, you can simply delete the has-definitional-manifestation relationship, and make the concept primitive if necessary, and allow further refinement of editorial policy guide the future disposition of those concepts. 

Recommendation

Detailed design final specification

Convert all the stated Has definitional manifestation relationships to Is a relationships except those identified in Section 6.3.3. Evaluation of Revised Design.
This conversion can be easily accomplished by using the batch editing functionality At this point, given that there are only a little less than 300 of these terms that has HAS DEFINITIONAL MANIFESTATION as a stated relationship, and the lack of knowledge of the overall impact of a global ISA change, I would not recommend batch processing. If you look back at my angle closue glaucoma example, it may be possible to greatly improve the overall inferences for these concepts with proper review and modeling of the related terms.
Again, I recommend testing this with a random sample of existing concepts (i.e 20-30) to determine the impact.How about if I have Monique review all my recommended changes, and either note agreement, or an alternative definition for the implementation phase person to then use to edit content? of the workbench, importing the file of stated Has definitional manifestation concepts into the list view. This file is located at https://csfe.aceworkspace.net/sf/go/doc12906?returnUrlKey=1440355469416This is no longer available – would it be possible to add the list as an appendix.The list view is no longer supported by the current IHTSDO tooling. The document "has definitional manifestation," recommendations for update, and identification of concurrent issues, has a complete list. {-}.-Since the original tool is no longer available, one would hope that the new tooling could at least search to locate all the concepts that contain has-definitional-manifestation, and then work from there. If the environment does no support such queries, then I hope the business requirements and implementation plan are updated to reflect the necessity of such a feature.

Iteration plan

The list in Section 6.3.3. Evaluation of Revised Design, can be iterated over as many times as necessary to get it right.

Quality program criteria

Quality metrics

I propose 2 metrics related to the Has definitional manifestation relationship evaluation.

Quality metric 1

Quality metric 2

Project Resource Estimates

Estimate project size; Editing of 292 concepts that will affect ~4000 concepts. The work of processing the particulars of these concepts, and presenting them in a document for review has already been done. A modeler should be able to accomplish a review of the data necessary to complete this task in 30 hours. Once the list is accepted, implementing the change would take less than 1 hour. If the list already provided is considered sufficient, implementing the change will take less than 1 hour.

Scope of construction phase

Skills required: A general ability to understand is-a relationships, and to understand the medical concepts that the Has definitional manifestation is currently applied to. An ability to edit content in the IHTSDO workbench, and to use the batch editing functional.
The elaboration phase has provided a number of reports, and a complete review of all concepts with stated use of Has definitional manifestation. Based on the review, a recommendation has been made for each concept, and other issues such as potential missed synonymy have been noted (See the doc of "has definitional manifestation," recommendations for update, and identification of concurrent issues. This and the links below have moved since we have moved over to Confluence. Either add as appendices or new reference.This document will be linked to this page in confluence. The index at the beginning of the document shows each of the concepts, and categorized the recommended solution for each concept. This document provides the stated and inferred relationships for each concept, so that those relationships can be conveniently reviewed.
In addition, a file suitable for importing into the list view of the workbench is provided (doc12906: https://csfe.aceworkspace.net/sf/go/doc12906?returnUrlKey=1440355469416), so that reviewers or implementers can easily focus on the desired concepts within the workbench. Doc12892 and doc12906 are both in alphabetical order to facilitate ease of review.
There is a place in Doc12892 for reviewer's to document their comments on each concept.
Implementation: A second modeler will review the changes to all the concepts with a relationship of Has definitional manifestation, and promote those that are satisfactorily modeled into the next SNOMED Release. Using the batch editing functions will allow quick conversion of has definitional manifestation to is-a relationship, if the proposed changes are accepted. I am happy to author these changes if a review indicates an acceptance of the approach, or identifies specific exclusions.
Preventing recurrence of problem: removing the Has definitional manifestation attribute from approved use, and replacing with appropriate is-a relationships will prevent recurrence of this problem.

Projection of remaining overall project resource requirements

Expected project resource requirement category

The project resource requirement is classed as SMALL – less than 1 person week. This project can be considered fast track, with minimal to no project management necessary.

Expected project impact and benefit

The project impact is MEDIUM – significant improvement to a minority but high profile use case

Indicative resource estimates for construction, transition and maintenance:

Implementation/Construction and transition phase: ~300 concepts to be given updated definitions
Maintenance phase:No new concepts are anticipated from the maintenance phase of removing Has definitional manifestation. There may be new concepts created as definitions are improved to replace the use of has definitional manifestation, such as addition of Angle closure obstruction (morphologic abnormality) to the morphology concepts, as suggested above.