Observational Health Data Sciences and Informatics

The most straightforward way to implement a local drug Vocabulary is to utilize RxNorm Extension logic to extend standard Drug concept pool and map source drug products to concepts in RxNorm extended hierarchy.

To incorporate a new set of drug information, a structure should be achieved that contains every Concept only once and preserves the RxNorm structure, no matter which vocabulary the additional Concept is coming from. In a way, it should create a mixed RxNorm/drug vocabularies union.

In order to achieve this, any two equivalent Concepts have to be matched through its components: Ingredients to Ingredients, Forms to Forms, Supplier to Supplier, etc. Concepts are defined as matching if all components match. For example, a Clinical Drug matches another Clinical Drug if it contains the same Ingredients at the same strength and the same Dose Form.

Standard concepts in Drug domain are placed in a single comprehensive hierarchy based on their attributes. To correctly implement a vocabulary in CDM, following attributes must be extracted:

1. Ingredients: active substance(-s) in pharmacological preparation. Examples: Aspirin, Trastuzumab, Ibuprofen etc.;
2. Dose form: form in which drug is administered to patient. Examples: Oral Tablet, Transdermal System, Injectable Suspension;
3. Marketed/Brand Name: Nurofen, Kadcyla, Prevenar 13 etc.;
4. Dosage/strength of active substance, total volume of liquid drug or number of actuations for dosed inhalers or applicators;
5. Information about combination pack contents, e.g. contraceptives calendar packs with varying dosages and ingredient combinations;
6. Box size: number of unique package units (like bottles, blisters or syringes) in a packaged box.
7. Manufacturer/supplier name (Reckitt Benckiser, TEVA, Ratiopharm etc). Generally, local branch and general name of international entity are considered same (“Teva Europe” is the same as “Teva”).

Such attributes may be given explicitly in a well-structured source vocabulary or have to be extracted from drug product names. In case of drug products or discrete attributes like Ingredients or Brand Names not having their own codes, new codes have to be constructed combining the word “OMOP” and a running number. The running number should be unique across all vocabularies. That means, each time a new vocabulary is added or refreshed, the next Concept Code should be the one of the last (without the 'OMOP' string) +1.

All source vocabulary concepts, extracted attributes and dosage and packaging information must be staged in standardized format of input tables, that can processed to be included into CDM Standardized Vocabularies.

To implement a tool to create and maintain the above structure, a number of issues need to be taken care of:

• Excipients: There is no general agreement of what is an active agreement and what is an excipient. Therefore, some of the ingredients need to be declared as “semiactive”, such as gelatine. Generally, if ingredient can be biologically active, but is not present in preparation for its pharmaceutical properties, it should be considered excipient. Excipients should be excluded from a list of a drug's ingredients.
• Forms: These are not used the same way across drug vocabularies. For example, RxNorm has a Form “Cream”, but also “Ophthalmic cream”, “Vaginal cream”, “Rectal cream”, “Oral cream” and “Cutaneous cream”, making this Form ambiguous. Instead of a one-to-one mapping, a one to many mapping with an order of precedence is required to establish matching equivalence between Forms.
• Strength: RxNorm normalizes weight units to “mg” and volume units to mL, but other vocabularies might not. There might be units like “µg”, “gram-%” or “volume-%”. Special unit conversion tables are needed instead of simple unit mappings. This approach becomes infeasible if units are used where the conversion is dependent on the molecule, like “mol” or “equivalent”.
• Ingredient forms: Ingredients might have ambiguous chemical forms, which RxNorm calls “Ingredient” and “Precise Ingredient” (e.g. a salt of the active compound). They have to be mapped to the right Standard RxNorm Ingredient. If there is no RxNorm Ingredient to map to and the drug vocabulary to be added contains several ambiguous forms of the same Ingredient, one of them has to be declared Standard. In rare cases there might be several Standard duplicates of the same Ingredient. In those cases mappings from source vocabularies must be made with precedence. Another problem might occur due to the strength is given for a precise ingredient, rather than a standard ingredient. An ingredient that is presented in the way of aqueous/spirit extract should be considered as the same one.

The new vocabulary should be prepared in the following tables:

This table is expected to contain as a minimum the comprehensive list of Concept Classes:

• Drug Product (Branded Drug, Clinical Drug, Marketed Product etc.)
• Form
• Brand Name
• Ingredient
• Supplier
• Device (for source conccepts falling outside of Drug cathegory)

It may contain Branded or Clinical Drug Forms or Components, but if not they will be derived (see below). Note that units should not have their own concept in the DRUG_CONCEPT_STAGE table. Instead, they should be used as verbatim. If the precise Concept Class is not known, it can be included as “Drug Product” and the correct Concept Class will be assigned during the incorporation automatically based on the availability of Strength, Dose Form, Brand Name, Supplier, Quantity and Box Size information.

Brand Names that are simple combinations of generic international name of active substance and manufacturer name (e.g. “Aspirin Bayer”) should not appear as attributes for Drug Products. Manufacturer information should be stored as a concept with Supplier class.

Concepts that belong to the source vocabulary, but do not belong to Drug domain by OMOP rules., should be classified as 'Device'. Typically, these belong to different substance groups:

• Radiographic contrasts and dyes (barium sulfate, radiolabeled iodides)
• Nutritional supplements without stated mechanism of pharmaceutical action (e.g. herbal complexes, homeopathic preparations)
• Formulas for infant feeding
• Parenteral nutrition (aminoacids and/or lipid mixes)
• Solution for dialysis, catheter maintenance etc.
• Blood products for transfusion, blood plasma, autologous and non-autologous transplants of any kind
• Cosmetics, sunscreens, non-medicated shampoos and soaps, etc.
• Surgical materials like bone cements
• Topical/external disinfectants

Animal drugs can be handled as Drugs or Devices, depending on what their role in patient data can be expected to be. Note that only concepts from Drug domain can have attributes.

This table should contain the mapping between source codes and Standard Concepts for Ingredients, Brand Names, Dose Forms, Suppliers and Units.It also may contain mapping from source drugs to Standard Concepts for related ATC classes. All other relationships will be ignored.

This table should contain all mappings from the new to existing Concepts and their precedence.

Units should be mapped to Standard Concept Units. Weight units should be converted to milligram, volume units should be mapped to milliliter, molar - to millimole with the right conversion factor. The source_code field should contain the verbatim string of the unit. It is highly desirable to only use units that are in use by Standard native RxNorm concepts. Querry DRUG_STRENGTH table for a distinct list.

Ingredients must be usually mapped to Standard concepts one to one. If ingredient is given as a mix (e.g. Co-dried gel of Magnesium Carbonate and Aluminium Hydroxide), it should be split in multiple enitities with distinct new codes; each component of the mix must be mapped to standard ingredient.

One to many mappings with precedence should be used if:

• Source ingredient is an ion (like calcium, iron, zinc, etc.), which should be mapped to all it's salts;
• Source ingredient is a herbal extract, which should be mapped to all suitable standard concepts;
• Target vocabularies contain duplicates of standard ingredients. This is rare.

Dose Forms are commonly mapped to multiple RxNorm dose forms with precedence. Modified release forms should be first mapped to corresponding forms in RxNorm vocabulary (like Delayed Release Oral Capsule), and then to more generic forms (Oral Capsule) with lower precedence.

This table should contain relationships for each Drug Concept: To the Ingredients (always), the Dose Form (if appropriate),the Supplier (if appropriate) and the Brand Name (if appropriate). All other relationships will be derived and ignored if they exist in the table. The relationships need not be symmetrical, only the one initiating from the Drug Concept is required.

If Drug Product concept does not have an Ingredient attribute, it will not have any standard mapping target after processing. Supplier attribute will not be considered for concepts without DS_STAGE or PC_STAGE entry since Marketed Product concepts can not exist without dosage information.

This table contains the dose of each ingredient in each drug, as well as the box_size. For drugs which have no strength information or have only for some of the containing ingredients, the ds_stage record must be omitted. '0' values in ds_stage are only allowed for inert drugs. Drug ingredients should match those in internal_relationship_stage. If ingredients are mapped to the same one in relationship_to_concept their dosages should be summed up. A drug should not contain ingredients in solid (amount) and liquid (numerator/denominator) form. This might be caused be either source data aberration or drug pack, which must be split into separate Drug Products and processed in PC_STAGE table.

• Inhalers, enemas or sprays that release certain dosage of active ingredient per activation should also be stored in numerator/denominator form with total number of actuations as denominator (e.g. X MG / Y ACTUAT). Drugs that release active substance over prolonged period of time, like transcutaneous patches or modified release oral pills, can also be stored as numerator/denominator with hours as denominators.
• All drugs with fixed amount must have dosage in amount fields and all solutions must have dosage filled in numerator and denominator fields. When liquid drugs in data contain concentration information without volume, DENOMINATOR_VALUE field is left empty.
• Gases for inhalation must be put in numerator fields with % in unit field without filling denominator fields. It’s the only acceptable use of percents in DS_STAGE. Make sure to convert everything else to mg/ml or mg/mg.
• Patches, drug implants and other forms that release molecules over a period of time (even extended release tablets or capsules) may also be stored in numerator/denominator form (e.g. X MG / Y HOUR).
• Inhalers, enemas or sprays that release certain dosage of active ingredient per activation should also be stored in numerator/denominator form with total number of actuations as denominator (e.g. X MG / Y ACTUAT).
• If dose form for the source concept is given as a soluble powder without a solvent (except powder inhalers), dosage is stored in amount field.
• For drugs that are administered in a form of oral liquid (solution, suspension, syrup), denominator value of 5 ML should be kept only when we are certain that the dosage is not given “per tbsp.”; if 5 ML is not an actual fixed administered dose (e.g. a sachet or vial), it should be treated as concentration (DENOMINATOR_VALUE = NULL).
• Box size equal to 1 should be simply stored as NULL.
• If source provides conflicted, mixed or incomplete info on one or more ingredients dosage or solution volume, drug should not have entries in ds_stage

This table contains the composition of a Clinical or Branded Pack: The Clinical or Branded Drug and, number of doses in each box and number of boxes in each pack. If it is a boxed Pack, it will also contain the box size, since Packs have no records in DS_STAGE like the other drug products. Packs are allowed to have branded drugs as components, although usually Brand Name is only attributed to packs as a whole. Supplier may only be atrributed to the pack as a whole.

This table contains alternative names for concepts. These are either alternative names provided by source or names in original languages, since DRUG_CONCEPT_STAGE will contain english names only.

CONCEPT_RELATIONSHIP_MANUAL

This table allows to manually map source concepts to existing standard concept in OMOP CDM circumventing standard vocabulary building process. This is useful to represent source concepts that belong to a different domain, preserve relationships inside the vocabulary or to map concepts to standard Drug concepts from outside of RxNorm and RxNorm Extension logic (e.g. standard concepts in ATC or CVX vocabulary.

This table needs not to be symmetrical like CONCEPT_RELATIONSHIP; complementary relationships will be built automatically. Note that concepts with equivalence mappings in this table should not have relations to attributes in other input tables.

The input tables need to have the following quality requirements:

For quality assurance of input tables you can use drug_stage_tables_QA.sql script from project's github.

All propositions to add a new vocabulary into CDM may be submitted (optionally with prepared input tables) as issues on github.

Matching This step is necessary to add inferred equivalence relationships between new and existing Standard Concepts. All matches are created.

The matching considers all components (Ingredient-strength(s), Dose Form, Brand Name, Quantity, Box size) in the order of precedence. A 10% mismatch between strength or total volume values is still considered a match. Matching beteween normal and Quantified products compares the Numerator Value of the non-quantied to the Numerator divided by the Denominator Value.

Inferring of missing Concept Classes

After RxNorm Extension Standard concept with the fullest attribute set will be created for each source entity, all missing preceding Concept Classes are inferred from the existing ones, from bottom upwards.

Even though all drug classes are inferred, only those will be written to the CONCEPT table that have no mapping to an existing equivalent Standard Concept.

Result All records in the DRUG_CONCEPT_STAGE table are written to the CONCEPT_STAGE table as follows. The standard_concept field is set to 'S' for all products and Ingredients and Brand Names that have no match to existing Standard Concepts. Dose Forms, Brand Names and Suppliers are always written as non-standard.

All records linking drug products to their Ingredients, Dose Forms, Suppliers and Brand Names are written to the CONCEPT_RELATIONSHIP_STAGE table. Note that this can be a one or two step connection:

* Ingredients, Dose Forms, Suppliers and Brand Names that have no equivalent to RxNorm (and are therefore Standard Concepts): These are converted from the INTERNAL_RELATIONSHIP_STAGE table. * Ingredients, Dose Forms and Brand Names that have an RxNorm equivalent (at least one) are not written into the CONCEPT_RELATIONSHIP_STAGE table, but the RxNorm equivalent instead, using the records from the RELATIONSHIP_TO_CONCEPT table with the relationship_id = 'Has standard ing', 'Has standard brand' and 'Has standard form' .

Relationships between Drug Products or derivatives (Drug Forms and Components) are connected through CONCEPT_RELATIONSHIP_STAGE records with the the following relationship_id values:

Name of the Branded and Clinical Pack concepts is built by combining names of the contents.

Relationships between any Drug Concept Class and a Classfication Concept Class is recorded through the “Drug has drug class” and “Drug class of drug” generic relationship pair.

Finally, a new DRUG_STRENGTH_STAGE table should be created from DS_STAGE and the the unit conversions in RELATIONSHIP_TO_CONCEPT, so the content can be added to the DRUG_STRENGTH table. This includes only Drug Concepts that have no mapping to an existing Standard Concept and are now Standard themselves.