How to access and query the New Mexico Decedent Image Database (NMDID)

At an RTI International Forensic Technology Center of Excellence webinar, Dr. Shamsi Berry, assistant professor of biomedical informatics at Western Michigan University Homer Stryker M.D. School of Medicine (WMed), demonstrated how to access and analyze metadata from the New Mexico Decedent Image Database (NMDID) and answered attendee questions on standards, download limits and medication coding.

Why it matters: The NMDID provides de-identified decedent records that researchers and public-health analysts can use to study mortality patterns, injury mechanisms and associated health conditions. Berry's walkthrough clarified which fields are searchable on the website, which require offline relational queries, and what standards (LOINC, SNOMED CT, NCI Thesaurus, RxNorm) the project uses to make data comparable across sources.

Berry said the database's data dictionary is organized into three sections'census, health and circumstances of death'and that each variable includes cardinality (1:1 or 1:many), a definition, its data source (medical examiner records, next-of-kin interview, or both) and a normative-answer list. "It's broken down into 3 sections, census data, health data, and circumstances of death," she said, adding that not every decedent has every field because entries depend on what was available in the medical examiner's file or next-of-kin information.

On standards and coding, Berry pointed attendees to LOINC for measurements, SNOMED CT for clinical terms, the NCI Thesaurus for oncology vocabularies and RxNorm (via the RxNav browser) for medications. "I used a natural language processing tool called Canary that was developed at Harvard and is freely available," she said, describing how the project used the tool to extract medication names from originally free-text fields.

Berry demonstrated a simple search (males, divorced, suicide as manner of death, with a diagnosis) that returned 93 matching records; after exporting and loading results into Microsoft Access, she showed a joined query that identified 82 of those 93 records with a recorded diagnosis of mental illness and that 12 of those were age 65 or older. "There are 93 individuals that fit this," she said. "Of those 93 individuals we had, 82 of them have a diagnosis of mental illness." She also used oxycodone as an example for medication coding, noting the RxCUI (RxNorm Concept Unique Identifier) for oxycodone is 7804 and that adding that drug number to the medications table returns matching records.

Practical steps Berry outlined: use the site's cart to gather results (the web UI pages 100 results at a time), download CSVs (the site limits a single download to 500 results), save each CSV as an Excel file and import the tables into Microsoft Access, then link tables by the de-identified record number (the primary join key) to run simple and complex queries. She warned that some fields remain incomplete'for example cadaver condition and bone density'and that medication extraction using Canary has caveats (the medication list may be out of date and misspelled drug names may not be captured). "The list hadn't been updated for a year or so," she said, describing the limits of automated extraction.

Resources and next steps: Berry noted the NMDID site includes training videos that demonstrate simple and complex Access queries and said new fields will be announced on the site when they become available. She said she is seeking additional funding to enable direct searchable fields for specific diagnoses to simplify end-user searches.

The webinar concluded with a short Q&A on standards, medication extraction, prevalence calculations and field availability; attendees were directed to the NMDID and WMed contact information and FTCOE resources posted in the webinar chat.