New: the AI scribe is live. Speak the consultation, review a coded draft. See Thravi’s AI
All posts

From uploaded lab reports to reliable patient trends

A pathologist comparing printed laboratory reports with patient trend charts on a monitor

A patient brings reports from three laboratories. One prints “Hb”, another “Hemoglobin” and another “Haemoglobin”. A doctor recognises the labels. Software needs enough context to establish what was measured before placing those results on the same trend.

This is a practical challenge for digital health: making records from different sources useful together. Uploading a PDF makes it accessible. Turning it into reliable, comparable observations takes more work.

OCR reads the page; normalization establishes meaning

Optical character recognition extracts text. A useful extraction workflow must also associate each result with its test name, unit, reference range, patient and report date. It should preserve qualifiers such as a result reported as “less than” a threshold.

Normalization then asks whether an extracted field corresponds to a known test. Similar names can be candidates for a match, but a name alone is insufficient. A panel heading such as “LFT” describes a group of tests; it should not be treated as an individual result.

Where LOINC helps

LOINC provides identifiers for observations. Its naming model distinguishes what is measured, the property, timing, specimen or system, result scale and, where relevant, method. Those distinctions explain why matching only an abbreviation can select the wrong code.

For example, a shared word such as “glucose” does not establish that two observations describe the same measurement. The surrounding report details matter. The official LOINC guide to the major parts of a term explains this model.

A useful trend needs more than a matching code

Even after mapping, software must check units, dates and duplicate imports before combining values. Any unit conversion should be explicit and retain the original value. Missing context should remain visible for review.

A reviewer should be able to open the source report from an extracted result, see what was changed and correct an uncertain match. A chart becomes much more useful when every point has a traceable source.

How this connects to ABDM

India's ABDM FHIR implementation guide defines structures for exchanging clinical records, including diagnostic reports and observations. Structured lab data can contribute to that exchange, but assigning a LOINC code alone does not establish ABDM conformance.

The wider exchange also requires the applicable record structure, patient context and consent workflow. See the NRCeS ABDM implementation guide and the Ministry of Health's explanation of consent-based ABDM exchange.

What we are working toward at Thravi

Thravi already supports consultation-linked lab orders and structured result entry. Our work on uploaded-report normalization explores how external documents can contribute to a longitudinal patient record, with source verification and review of uncertain mappings. This is ongoing development, rather than a claim of fully automated report interpretation.

For hospitals evaluating this capability, ask a vendor to demonstrate a difficult report: mixed units, ambiguous labels and a repeated upload. Inspect how the system handles uncertainty as carefully as the final chart.

Read our guide to clinical coding standards or explore Thravi's EMR software to see how structured records fit into the wider workflow.

Book a personalised demo

Tell us how your facility operates, and we’ll show you how Thravi can be configured for your setup.