How TranslateMD Translates Between Healthcare Systems

When a German patient brings their discharge summary to a US cardiologist, there are two translation problems — and most services only solve one.

The obvious problem is language: German to English. The harder problem is that the document encodes clinical information in Germany’s healthcare system. The diagnosis codes are ICD-10-GM (German Modification), not ICD-10-CM. The procedures are OPS codes. The medications are identified by PZN numbers and German brand names. The hypertension is coded as I10.00 (no crisis) — a subclassification that doesn’t exist in ICD-10-CM.

A fluent German-speaking doctor can read the words. But without system-level translation, the codes are either invisible or wrong.

TranslateMD solves both problems. Here’s how.

How TranslateMD Processes a Document

Every document that enters TranslateMD goes through structured processing designed for medical records specifically.

Parsing and Structure Detection

TranslateMD accepts structured and unstructured medical documents: discharge summaries, lab reports, prescription records, dental records, referral letters. Documents arrive as PDF, DOCX, or plain text.

The engine identifies document type, sections, and language before the content is touched. This matters because different document types carry different code types. A dental record requires dental notation mapping. A discharge summary requires diagnosis and procedure code mapping. A prescription requires drug name and scheduling translation.

Healthcare System Identification

TranslateMD identifies which coding systems are present in the document:

ICD-10 variants: GM (Germany), CM (US), TM (Thailand), WHO-based (Mexico, India, Turkey)
Dental notation: FDI (most of the world), Universal/ADA (US, Canada), Palmer (UK, some EU)
Drug identifiers: ATC codes, PZN (Germany), NDC (US), brand names in each country
Procedure codes: OPS (Germany), CPT (US), BEMA/GOZ (German dental)

This identification uses country context — if the document origin is Germany, the engine looks for GM-specific features like the exclamation mark suffix (!) on secondary codes, the I10.0x/I10.1x hypertension subclassification, and U50-U52 functional disability scores that have no WHO or CM equivalent.

Clinical-Aware Language Translation

This is where the linguistic translation happens — but it’s not a call to a generic translation API. TranslateMD uses clinically-aware translation that preserves:

Clinical negation (“no evidence of” must remain negative)
Anatomical precision (laterality, severity gradations)
Temporal markers (acute vs. chronic, onset dates)
Uncertainty language (“suspected”, “rule out”, “probable”)

Generic translation models frequently collapse these distinctions. “Verdacht auf” (German for “suspected”) has important clinical meaning. A document that says “Verdacht auf Pulmonary embolism” must not emerge as “Pulmonary embolism.” That’s not a translation error — it’s a diagnostic distortion.

Medical Code Mapping

The code mapping is the core technical differentiation of TranslateMD. It maps detected codes across healthcare systems using a validated knowledge base — not AI inference.

Here’s a concrete example. A German document contains the code K02.1. In ICD-10-GM, this is “Dentinkaries” (dentinal caries). In ICD-10-CM, the K02 category was reorganized entirely. There is no direct K02.1 equivalent. The correct US mapping depends on the surface involved:

K02.51 — Dental caries on pit and fissure surface limited to enamel
K02.52 — Dental caries on pit and fissure surface penetrating into dentin
K02.61 — Dental caries on smooth surface limited to enamel
K02.62 — Dental caries on smooth surface penetrating into dentin

Without context from the record (X-rays, clinical notes), K02.9 (unspecified dental caries) is the safest approximate. TranslateMD flags this with an annotation: the source code was K02.1, the target code is K02.9 (approximate), and a clinician review is recommended before billing.

Another example: German hypertension codes I10.00 and I10.10. The GM modification sub-classifies essential hypertension by crisis status (I10.00 = no crisis, I10.01 = with crisis) and malignant vs. benign type (I10.1x = malignant). ICD-10-CM dropped both distinctions decades ago — all essential hypertension is a single code I10. TranslateMD maps both GM variants to I10 and annotates that clinical nuance was lost in translation.

Clinical Annotation

The output is wrapped with structured clinical context:

Source system metadata: which coding system the document used, country of origin, system version
Mapping confidence indicators: exact match, approximate match, unmapped
Safety flags: drugs with different regulatory status, codes with no equivalent, collapsed specificity
Reviewer prompts: suggested questions a receiving clinician should ask

The output is a translated document that a clinician can read with confidence — and with full visibility into where the translation was exact and where it was approximate.

Knowledge Base vs. AI Inference

Every code mapping in TranslateMD comes from a curated, validated knowledge base. No AI inference required for mapped codes.

The knowledge base covers:

Thousands of validated code mappings across ICD-10 variants, FDI/Universal/Palmer dental notation, OPS, CPT, BEMA, GOZ, CDT, ATC, NDC, and PZN
11 country profiles with complete healthcare system context (US, DE, MX, IN, TH, TR, GB, KR, AE, BR, JP)
Comprehensive reference documentation covering per-country system guides, cross-system crosswalks, notation systems, and medical terminology

In our evaluation suite — 46 test cases covering all Tier 1 and Tier 2 corridors — TranslateMD achieves:

100% accuracy on standard cases (well-documented, high-frequency code pairs)
94.5% accuracy on edge cases (rare codes, partial matches, system-specific features)

For the small percentage of edge cases where no validated mapping exists, TranslateMD falls back to AI-augmented mapping. The AI fallback is transparent — the output is clearly marked as AI-proposed rather than knowledge-base-confirmed.

Why System-Level Translation Matters

The consequences of system-mistranslation are clinical, financial, and administrative.

Clinical: A drug that is a common OTC painkiller in Germany — metamizole (Novalgin) — was banned by the FDA in the US in 1977 due to risk of agranulocytosis. A US pharmacist reading a German prescription with Novalgin needs to know immediately that this drug is not available and not safe by US standards. A word-for-word translation delivers the brand name. TranslateMD delivers a safety alert.

Financial: Insurance reimbursement in the US runs on ICD-10-CM. A German code submitted without conversion will be rejected. Worse, if the code is converted incorrectly — K05.3 (chronic periodontitis, GM) submitted as K05.3 instead of K05.31 (localized) or K05.32 (generalized) — the claim may be underpaid or trigger an audit.

Administrative: Many cross-border health systems (EU patient mobility, international student health insurance, medical tourism) require proper code translation for claims processing. Incomplete or approximate translations create administrative burdens that fall on patients.

What TranslateMD Does Not Do

TranslateMD is not a medical decision-support system. It doesn’t diagnose, recommend treatment, or assess clinical appropriateness. It translates the clinical record that exists.

The annotations and reviewer prompts are tools for the clinician — not replacements for clinical judgment. When TranslateMD flags a mapping as approximate, it’s telling the receiving clinician: “here is the best available code given the source document; verify against the original record before acting.”

Get Started

TranslateMD is available via API and web interface. The API supports single-document translation and batch processing with the --batch flag for high-volume workflows.

For teams handling regular cross-border patient records — medical tourism coordinators, international health insurance processors, border-region health systems — TranslateMD Pro and Team plans include full corridor coverage, safety alert notifications, and dedicated support.