ATC and ICD10 are often discussed together, but they classify different things. ATC codes describe medicines—their anatomical target, therapeutic use, pharmacology, and chemical substance. ICD10 codes describe diseases, symptoms, injuries, and health conditions. Used together, they help connect what was prescribed with why it was prescribed, without treating the relationship as a fixed one-to-one mapping.
For Indian hospitals, pharmacies, insurers, health-tech teams, and researchers, that distinction matters. A reliable coding workflow improves data quality, auditability, reporting, and clinical interpretation. A careless workflow can create incorrect claims, misleading dashboards, or unsafe AI outputs.
ATC and ICD10: What Each System Represents
The Anatomical Therapeutic Chemical (ATC) classification is maintained by the World Health Organization Collaborating Centre for Drug Statistics Methodology. It groups medicines into five levels:
- Level 1: Anatomical main group—the organ or system affected, such as the cardiovascular system.
- Level 2: Therapeutic subgroup—the broad therapeutic purpose.
- Level 3: Pharmacological subgroup—the drug’s pharmacological action.
- Level 4: Chemical subgroup—a narrower chemical or pharmacological grouping.
- Level 5: Chemical substance—the specific active ingredient.
An ATC code identifies a medicinal product or active substance in a standardised hierarchy. It does not, by itself, state that a patient has a particular disease or that the medicine was appropriate for that patient.
ICD10, meanwhile, classifies health conditions. Depending on the jurisdiction and implementation, codes may represent diagnoses, symptoms, external causes, complications, or factors influencing health status. India-based teams should confirm which ICD-10 release, national adaptation, payer rule, or hospital coding policy applies to their use case.
A prescription for metformin, for example, may be represented with an ATC code for the medicine, while the patient’s documented condition may receive an ICD10 code for type 2 diabetes. The prescription does not prove the diagnosis: medicines can be used off-label, for prevention, for comorbidities, or for reasons not captured in a single encounter code.
Why the Pairing Is Useful
Linking medication and diagnosis data can support:
- Clinical review: clinicians can examine treatment patterns alongside documented conditions.
- Pharmacy analytics: teams can track utilisation by active ingredient, therapeutic class, or patient population.
- Claims and reimbursement: diagnosis and procedure documentation can be checked against payer requirements, without assuming that an ATC code replaces a diagnosis code.
- Public-health research: researchers can study prescribing trends, treatment gaps, and medicine access.
- Hospital operations: procurement and inventory planning can use aggregated medicine classes and demand patterns.
- Healthcare AI: models can use medication and diagnosis features for cohort selection, risk analysis, or research—provided the data is de-identified and clinically validated.
Teams preparing datasets for machine learning should also review ICD-10 codes for LLM training. The same principles apply: preserve code-system provenance, document transformations, and prevent a model from inferring facts that are not supported by the record.
A Safe Workflow for Using ATC ICD10 Codes
1. Define the business or clinical question
Start with the question, not the code table. Are you measuring antibiotic use, identifying patients with a documented condition, validating a claim, or building a research cohort? Each goal requires different inclusion criteria and may need additional fields such as encounter type, age, dosage, route, or prescription status.
2. Identify the exact code systems and versions
Record whether the medicine data uses ATC, a national drug formulary, an internal product code, or a combination. Record the ICD-10 version and local modifications. Do not silently mix WHO ICD-10, ICD-10-CM, ICD-10-AM, or other adaptations; similar labels can have different definitions and levels of detail.
3. Normalise medicine records
Standardise active ingredient, strength, formulation, route, and quantity. A brand name alone is insufficient for dependable analysis. Combination products require particular care because one product may contain multiple active ingredients and have more than one relevant classification.
4. Preserve the original clinical evidence
Store the source diagnosis text, code, prescription, encounter date, and coding status where permitted. Keep any mapping table separate from the original record. This makes corrections, audits, and model evaluation possible.
5. Use mappings as analytical aids, not clinical conclusions
There is rarely a universal one-to-one relationship between an ATC code and an ICD10 code. Use curated mappings, inclusion and exclusion rules, and expert review. If a rule is probabilistic or intended only for cohort discovery, label it clearly.
6. Validate with clinicians and coders
Review samples across specialties, age groups, inpatient and outpatient settings, and common comorbidities. Check false positives and false negatives, not only overall match rates. For billing workflows, follow the applicable Indian payer, institutional, and regulatory requirements rather than relying on a generic internet code list.
Common Errors to Avoid
- Calling ATC and ICD10 a single coding system: They are complementary classifications with different purposes.
- Using a medicine to infer a diagnosis: The same drug may treat multiple conditions.
- Ignoring versions and update dates: Code definitions, classifications, and local policies change.
- Confusing product and ingredient data: Brand, salt, strength, and combination-product fields should not be collapsed without a documented rule.
- Mapping at excessive precision: A broad therapeutic class may not justify a highly specific diagnosis.
- Treating coded data as ground truth: Missing, copied-forward, or administrative diagnoses can distort analysis.
- Exposing identifiable health information: Apply access controls, minimisation, consent requirements, and applicable Indian privacy obligations.
Implementation Checklist for Indian Healthcare Teams
Before production use, confirm that your pipeline:
- Names the ATC and ICD-10 versions and source authorities.
- Stores code descriptions, effective dates, and mapping provenance.
- Separates clinical documentation from inferred or derived fields.
- Handles Indian brand names, generic names, combination products, and spelling variations.
- Supports human review for ambiguous records.
- Logs changes to mapping tables and model outputs.
- Tests performance across hospitals, languages, specialties, and care settings.
- Uses de-identified or appropriately governed data for analytics and AI development.
A lightweight rules engine may be enough for a reporting dashboard. A clinical decision-support or claims product needs stronger governance, testing, monitoring, and escalation. Builders evaluating implementation options can apply the same disciplined approach described in building your first machine learning app, while keeping clinical safety and validation ahead of model novelty.
Frequently Asked Questions
Are ATC and ICD10 the same?
No. ATC classifies medicines; ICD10 classifies diseases and other health conditions. They can be linked in a dataset, but neither replaces the other.
Can one ATC code map to one ICD10 code?
Usually not. A medicine may be used for several conditions, and a condition may be treated with several medicines. Any mapping should state its purpose and limitations.
Which system should a hospital use for billing?
That depends on the service, payer, contract, and applicable Indian coding rules. Confirm requirements with the organisation’s certified coding and billing teams rather than assuming ATC codes satisfy diagnosis or reimbursement documentation.
Can these codes be used to train healthcare AI?
Yes, but only with documented versions, appropriate governance, de-identification, expert validation, and careful separation of observed facts from inferred labels. For AI projects, also control API and infrastructure spending using a deliberate AI API cost strategy.
Apply for AI Grants India
Are you building responsible healthcare AI in India? Explore funding opportunities for Indian founders and research teams at AI Grants India.