OHDSI Home | Forums | Wiki | Github

Phenotype submission - Bodily injury

Cohort Definition Name: Bodily injury

Contributor Names: Anastasiya Nestsiarovich; Stuart R. Nelson; Nathaniel G. Hurwitz (deceased); Ruchina Shakya; Christophe G. Lambert

Outcomes: Moderate to high probability that bodily injury occurred to the patient, as inferred from procedures and diagnoses.

Overview:

Bodily injury, regardless of cause is a useful, albeit very large, basket of terms to capture that a person has experienced physical damage or trauma. Many conditions may contribute to the risk of injury. For example physical frailty associated with aging may lead to falls. Mental health conditions such as bipolar disorders could lead to thrill-seeking behavior that could lead to injury. Mental health conditions such as major depressive disorder and bipolar disorders can also lead to self-inflicted injuries. Occupational hazards can inrease risk of injury.

While there is generally clarity on which conditions represent injuries, and they are neatly categorized within the ICD-9 and ICD-10 hierarchies, this is not the case for procedures. It can be inferred from many procedures (e.g. bone repair), that an injury must have occurred. Other times one can infer, for example, with tumor removal, that an injury did not occur. However, many surgical procedures have ambiguity.

This phenotype development effort sought first, through multiple clinician review, to evaluate all of the procedures in CPT4, HCPCS, ICD9Proc, and ICD10PCS for which one can infer that it is moderately to highly probable that the patient experienced some form of injury from an external cause. In addition, because diagnosis codes for injury are straightforward to pull from ICD-9 and ICD-10 hierarchies we include these.

Approach

Note that the following procedure was applied in 2017, and may miss new procedure codes that have been added since then.

  1. A SQL query was generated to include and exclude candidate procedure terms in CPT4, HCPCS, ICD9Proc and ICD10PCS as follows:

    • Terms related to the following were included:

      • external cause of tissue damage: injury, trauma, laceration, fracture, dislocation, burn, epiphysis separation, disrupted, tear, wound, hematoma, haemorrhage, rupture, pneumothorax, lung hernia.
      • possible fracture/dislocation: cast, splint, brace, bandage, orthosis, fixation, immobilization, traction apparatuses, uniplane/multiplane application, vertebrae fusion, spinal fixation, posterior segmental instrumentation, cranial bones reconstruction, skeleton x-ray.
      • injury-related treatment strategies: debridement, sequestrectomy, graft, flap, skin transplantation, reattachment, replantation, decompression fasciotomy, repair of cerebrospinal fluid leak, decompressive craniotomy/orbitotomy, epistaxis (nasal packing).
      • damage caused by objects (weapon/drowning/swallowing): foreign body removal from organ walls/parenchima, soft tissues, orbit, brain, eye, organ lumens (pharynx, esophagus, larynx, trachea, bronchi), repair and matter extirpation from mouth structures (lips, buccal mucosa, gingivae, tongue, mouth floor, sublingual glands, soft and hard palate, uvula, epiglottis).
      • poisoning: antidote, emesis, gastric aspiration, toxicology.
      • repair of organs vulnerable to external influences: lips, mouth structures, external ear, nose, facial bones, extremities and body protective carcass: muscles/tendons/ligaments/fascias/nerves/capsulae/meniscus.
    • Terms that arose above but related to the following were excluded: congenital abnormalities, newborn/infants, neurological conditions (ptosis, paralysis), cardiac valves and vessels, dermatological lesions (except for burns and skin transplantation), neoplasms, obstetrics, perioperative states, aneurysms, ulcers, inflammatory and infectious diseases, arthrodesis, conization, hymen, circumcision, duct stones, diagnostics, plastic surgery, therapeutic fractures, incontinence, glaucoma, varices, placement of electronical devices, nonunion, malunion, immunization, radiation therapy, recurrent and persistent pathologies, ingrown nails, polyps, hemorrhoids, dental and orthodontic procedures, nasal turbinates.

  2. One clinician (Nestsiarovich) manually reviewed the list of included codes and made the necessary adjustments (for example, only bony structures radiography was selected; manipulations related to dentistry/orthodontics, telehealth, home care, extraocular muscles, small-sized structured (para-aortic body, glomus, pineal body, lacrimal gland), nasal turbinate were excluded, as well as codes containing both - inclusion term and term indicating its non-traumatic origin).

  3. The rest of the procedure list was reviewed by 3 clinicians (Nestsiarovich, Nelson, Huritz) and the final set of codes was defined based on their mutual agreement on “highly probable injury and moderate probable injury”, see below.

  4. Diagnoses related to non-medical injury
    We excluded adverse effects from substances (ICD9CM), iatrogenic injuries/consequences (ICD9CM and ICD10CM). We included all the remaining items under “injury” and ”external causes of morbidity” categories and about 300 additional items from other categories to select the following source concept codes appropriate for US data sources

    • ICD9CM codes: 8[0-9]*, 9[0-8]*, 99[0-5]*, E[0-8]*, E9[0-4]*, E9[6-9]*
    • ICD10CM include the following: S*, T*, V*, W*, X[0-5]*, X[9][2-9]*, Y[2-3]*

    The following PostgreSQL code specifies the ICD9CM and ICD10 diagnosis codes we used, following by a mapping from those to SNOMED:

create temporary table icd_injury as
SELECT concept_id from CONCEPT where vocabulary_id='ICD9CM' and concept_code ~ '^(8[0-9].*|9[0-8].*|99[0-5].*|E[0-8].*|E9[0-4].*|E9[6-9].*)'
UNION
SELECT concept_id from CONCEPT where vocabulary_id='ICD10CM' and concept_code ~ '^(S.*|T.*|V.*|W.*|X[0-5].*|X[9][2-9].*|Y[2-3].*)';

\copy (select * from icd_injury) to './injury_icd.txt'

create temporary table snomed_injury as
SELECT DISTINCT(concept_id_2) 
FROM concept_relationship 
WHERE relationship_id = 'Maps to' 
AND concept_id_1 IN (select * from icd_injury);

\copy (select * from snomed_injury) to './injury_snomed.txt'

The curated procedure codes were produced using the following criteria for categorization. High and moderate confidence procedures were retained for the concept set:

High confidence:

  • Terms indicating external cause of tissue damage: injury, trauma, laceration, fracture, dislocation, burn, epiphysis separation, disrupted, tear; to a lesser extent: wound, hematoma, haemorrhage, rupture, pneumothorax, lung hernia.
  • Terms indicating possible fracture/dislocation: cast, splint, brace, bandage, osthosis, fixation, immobilization, traction apparatuses, uniplane/multiplane application, vertebrae fusion, spinal fixation, posterior segmental instrumentation, cranial bones reconstruction, skeleton/bones x-ray.
  • Terms indicating injury-related treatment strategies: debridement, sequestrectomy, graft, flap, reattachment, replantation, decompression fasciotomy, repair of cerebrospinal fluid leak, decompressive craniotomy/orbitotomy, epistaxis (packing).
  • Terms indicating damage caused by objects (weapon/drowning/swallowing): foreign body removal from organ walls/parenchima, soft tissues, orbit, brain, eye, organ lumens (pharynx, esophagus, larynx, trachea, bronchi), repair and matter extirpation from mouth structures (lips, buccal mucosa, gingivae, tongue, mouth floor, sublingual glands, soft and hard palate, uvula, epiglottis).
  • Terms indicating poisoning: antidote, emesis, gastric aspiration, toxicology.
  • Terms indicating repair of organs vulnerable to external influences: lips, mouth structures, external ear, nose, facial bones, extremities and body protective carcass: muscles/tendons/ligaments/fascias/nerves/capsulae/meniscus.

Moderate confidence:

  • Procedures where injury and noninjury items are combined into the same term, for example: hematoma + abscess/seroma/cyst/bulla; traumatic disruption + aneurism/ulcer/diverticulum; excision of open wound + burn/scar.
  • Procedures where it is impossible to separate acute and chronic damages (dislocations).

Non-traumatic origin is probable:

  • skin/soft tissues/organ/eye lesions (might be dermatological disorders, polyps, neoplasms, etc.)
  • fistula (might be congenital if not determined as “traumatic”)
  • hernia (might be acquired or congenital if not determined as “traumatic”)
  • rupture of organs and vessels (might be due to obstetrical complications or internal pressure due to liquid accumulation, or aneurism)
  • amputation;
  • nail debridement/avulsion/repair (might be due to dermatological disorders);
  • eyelids reconstruction/repair (might be due to adenoma, entropion, ectropion, ptosis, chalazión, xanthelasma, nevi, etc.);
  • nasal septum reconstruction/repair (might be due to deviation or perforation);
  • breast reconstruction/implant/repair (might be due to mastectomy or lumpectomy);
  • vertebral corpectomy, autograft for spine surgery, bone graft (might be used for nerve decompression);
  • arthrotomy with cartilage repair (might be due to chronic cartilage destruction);
  • arthroscopy for debridement/chondroplasty/repair of cuff, labrum, lesion;
  • organ repair/resection/reattachment (might be due to surgery techniques or any disease like tumor/autoimmune etc.);
  • splenectomy (might be due to lymphomas, autoimmune diseases);
  • tendon/muscle/ligament transfer/graft/reposition (might be for strengthening the joint);
  • vessels repair/suture/ligation/occlusion;
  • repair (including plastic) of organs/joints/brain/eye;
  • esophagus dilatation (might be due to peptic stricture, achalasia, scleroderma, etc.)
  • blood transfusion;
  • deferred wound-healing (negative pressure wound therapy, collagen fillers, bioskin, etc.) (might be for trophic/complicated/non healing wounds, not acute traumas);
  • orbit exenteration; ocular implant; ocular surface reconstruction;
  • debridement of mastoid cavity;
  • omental flap;
  • tympanic membrane repair;
  • epistaxis with vessels ligation (might be due to complicated/persistent/uncontrolled by packing)
  • uterus laceration;
  • disc repair
  • Procedures with foreign body removal where foreign body might have endogenous origin or be self-introduced with no life-threatening consequences: joint cavities (cartilage), pleural/pericardial cavity and vessels (clots, emboli), ducts (calculi), digestive tract (except for pharynx), external auditory canal, nose etc.
  • Injuries/procedures with no specification of the nature of damage/intervention target: repair of body regions and organs, craniotomy with no aim indication, etc.

Logic description:

We provide three concept sets (lists of OMOP concept_ids):

  1. Procedures for moderate to highly probable injuries
    injury_procedures.txt (117.9 KB)

  2. Diagnoses for injuries – SNOMED
    injury_snomed.txt (36.3 KB)

  3. Diagnoses for injuries – ICD9CM and ICD10CM
    injury_icd.txt (561.4 KB)

Note that 1 and 2 are standard vocabularies, and 3 is non-standard. A cohort based only on standard vocabularies would use 1 and 2 only. Note: we only included mapped SNOMED codes for the ICD9 and ICD10 codes for injuries, and there are likely to be additional SNOMED codes for injuries that are not included here. Importantly we did not curate SNOMED procedure codes.

Assertion Statement: Simple queries determining the presence of at least one code in the given concept set were executed and at least one real person-level observational health data source and resulted in a cohort with at least 1 person.

Issues: The clinician-led procedure curation effort was performed in 2017, and new injury terms may have been added to the CPT4, HCPCS, and ICD10PCS vocabularies since then. Judgement calls of the likelihood of injury were made by three clinicians with disputes resolved by consensus. Others may disagree with these judgment calls, and they are open to revision. See also the notes under the Logic description section, especially that SNOMED has not been curated for procedures – we have only mapped ICD9CM and ICD10CM diagnoses to SNOMED (with the 3/31/2022 version of the vocabularies).

Thanks @Christophe_Lambert for your thoughtful post here. Just to clarify one procedural thing for you and the general community. The Phenotype Library is a community resource that contains cohort definitions, not just codelists. Now certainly codelists, which can be implemented as conceptset expressions, are an important building block to producing a cohort definition, and if building your conceptsets in ATLAS, you can embed your conceptset directly within the ATLAS cohort definition logic (e.g. you’ll see your conceptset in your JSON). But a codelist alone does not specify the domain and temporal logic about how to define the persons who qualify, it does not tell us the entry event for which a person enters a cohort (e.g. you need to couple a conceptset with a data domain, like 'a condition occurrence containing a standard concept of ‘bodily injury’), the inclusion criteria which are applied with temporal logic to qualify the entry events (e.g. you must have 0 condition occurrences of ‘bodily injury’ in the 60 days prior to the index date, to avoid double-counting a follow-up/prevalent code with an incident occurrence), or the cohort exit logic (e.g. the person belongs to the cohort for some duration of time, or continues in the cohort until some censoring event occurs or their end of observation period). I think the codelists you’ve generated could easily be incorporated into a cohort definition (albeit with the important caution that since much of your work is source code based, you need to ensure you use the codelists correctly to look into the source_concept_id fields and not the standard_concept_id fields, and with the full recognition that these definitions will not be appropriate for data sources that are not limited to your source vocabularies of interest). Perhaps you or someone else in the community would like to take on this work to build on these codelists to create the cohorts so that they can be included in the Phenotype Library?

@Gowtham_Rao has proposed Phenotype Library submission guidelines here: Cohort Definition Submission Requirements • PhenotypeLibrary.

Thanks, @Patrick_Ryan – we were working on this but got stuck with figuring out how best to tool the cohort for the HowOften effort, so figured, given the limited detail on structuring outcomes in the protocol that @hripcsa posted, that we might be figuring this out during the conference workshop.

Working on trying to generate a cohort today, I found that when building a cohort with this many codes Atlas also keeps dying in my web browser, first with “This page isn’t responding”, then “Aw, Snap! Something went wrong while displaying this webpage. Error code: SIGILL”. Even incrementally adding one rule at a time and saving, I eventually just could not make forward progress as it crashed every time.

I tried to attach the JSON cohort that I could get to work with standard concepts, but it was 11MB, exceeding the 4MB limit of this site. So I uploaded the human-readable version, trimmed down, here:
injury_nonstandard_human.txt (2.1 KB)

The beginnings of the source concept version was 47MB that I could not get any further on. There are 68,214 bodily injury ICD9CM/ICD10CM diagnostic codes whose source concept_ids map to either condition_occurrence or observation tables, and 15,028 procedures from the 4 procedure vocabularies (CPT4, HCPCS, ICD9Proc, ICD10PCS) that we curated.

One strange thing was that even though these 4 procedure vocabularies are considered standard, many of the codes from them do not have the ‘S’ denotation, but have NULL or ‘C’. Thus I was not entirely happy with using them with their concept_ids in the procedure_occurrence table.

Any suggestions?

Here is a Dropbox link to the JSON for our cohort using standard vocabularies, that we hope is suitable for loading into the phenotype library. We fixed an issue that there are non-standard terms among HCPCS, CPT4, and ICD9Proc, and used the concept_relationship mappings to map those to their standard terms (SNOMED procedures). Bodily Injury standard vocabulary JSON cohort definition

The SQL code was applied to the earlier injury_procedures.txt list:

create temporary table injury_procedures_standard as select distinct(concept_id_2) from concept_relationship where concept_id_1 in (select concept_id from injury_procedures) and relationship_id='Maps to';

\copy (select * from injury_procedures_standard) to './injury_procedures_standard.txt'

By upping the timeout settings on a Firefox browser while using Atlas, I was able to create a JSON version of our cohort using source vocabularies: ICD9CM, ICD10CM, CPT4, HCPCS, ICD9Proc, and ICD10PCS using source_concept_ids. It is a 47.8MB beast: Bodily injury source vocabulary JSON cohort definition .

It might be a nice stress test for the machinery. When editing cohort definitions, the listing of every code in a concept set within the GUI seems unnecessary and is probably what bogs the system down.

t