A hospital sees patterns before public agencies publish them. A rise in respiratory visits, a change in the age at which a condition is presenting, a cluster of road-traffic injuries on one corridor: all of it sits in admission and OPD records. Journalists want exactly this kind of local, current evidence, and most hospitals never offer it because nobody is sure how to release it without exposing patients.
This article is for hospital communications leads, medical records and quality heads, and data protection officers who want to produce data-led stories. It sets out a pipeline from question to release, specific aggregation and suppression rules, a verification routine and the legal points that matter. The depth is deliberately technical. The thresholds shown are working rules we suggest, not legal requirements, and the examples are hypothetical.
None of this is legal advice. Have your counsel and your data protection lead review the method before the first release, and have a clinician review every clinical statement.
What the law says, in plain terms
The Digital Personal Data Protection Act, 2023 defines personal data as any data about an individual who is identifiable by or in relation to that data (section 2(t)). It applies to the processing of digital personal data collected in digital form, or collected in non-digital form and digitised later (section 3).
Three consequences follow for data PR.
First, a published statistic that cannot be linked to any individual is not personal data. The work lies in making sure it really cannot be linked, and that is a technical question, not a labelling one.
Second, the processing that produces the statistic is processing of personal data. The admissions records are personal data. Using them for a communications purpose is a use that patients did not obviously consent to when they were admitted for care. Your data protection lead needs to decide the lawful basis and document it before analysts pull a single table.
Third, section 17(2)(b) exempts processing necessary for research, archiving or statistical purposes where the data is not used for decisions about a specific individual and the processing follows prescribed standards. Do not assume a PR analysis qualifies. A communications use is not the same as a statistical or research purpose, and the standards are set out in rules. Treat the exemption as a question for counsel, not as a safe harbour.
The Rules were notified in November 2025 with an 18-month phased compliance timeline, according to the Press Information Bureau. Build to the standard now, because the obligations are coming and because the NMC has already tied advertising to them: clause 7.6 of its October 2026 guidelines says advertisements involving digital personal data must comply with the IT Act and the DPDP Act and Rules, to the extent in force. Clause 7.4 says practitioners and clinical establishments must not share patient data such as names, faces or anomalies that can be identified publicly. Our existing note on DPDP and healthcare marketing covers consent in marketing generally; this article goes narrower, into release mechanics.
The pipeline, step by step
1. Start with the question, not the dataset
Write the story question first: "Are respiratory presentations in children rising earlier this season than last?" A precise question tells analysts which fields they need and, as importantly, which they do not. Data minimisation begins here.
2. Define the minimum data extract
Request only the fields the question needs. A hypothetical extract for the respiratory question might look like this.
| Source field | Keep as | Reason |
|---|---|---|
| Patient name, ID, phone | Drop | Never needed |
| Date of admission | Week start date | Reduces uniqueness |
| Date of birth | Age band (0 to 4, 5 to 12, 13 to 17, and so on) | Reduces uniqueness |
| Full address or pincode | City or district only | Pincodes can be small |
| Diagnosis code | Grouped category agreed with a clinician | Prevents rare-code identification |
| Free-text notes | Drop | Free text leaks identifiers |
| Ward, department | Department group | Fine detail adds nothing |
The extract should be produced by a named person in medical records or analytics, not by the communications team. Communications receives only the aggregated table.
3. Aggregate, then suppress small numbers
Aggregate to counts by period, age band and category. Then apply suppression: any cell with fewer than a set number of patients is merged into a wider category or removed. A floor somewhere between 5 and 10 is a common starting point in published health statistics. For a single hospital whose catchment is easy to guess, we suggest 10 as a starting point and a higher number for sensitive categories such as mental health, sexual health, paediatric oncology or any condition with a handful of cases a year.
Watch for two failure modes:
- Complementary disclosure. If you suppress one cell but publish the row total and the other cells, a reader can subtract and recover it. Suppress a second cell, or publish only broader totals.
- Differencing over time. Two releases a month apart that differ by one patient can identify that patient to anyone who knows when they were admitted. Release comparable figures at stable intervals and avoid publishing running totals that change by single digits.
4. Test for re-identification
Ask three people who know the community, none of them analysts, to try to identify an individual from the table and the surrounding story. Include the story text in the test: a quote saying "a 34-year-old teacher from the north of the city" combined with a small table can identify a person even when each piece alone is safe. Remove any anecdote that adds identifying colour.
5. Verify the numbers
Journalists will check, and so should you. A verification routine:
- Define each metric in one sentence in a methods note, including inclusion and exclusion rules and the date range.
- Re-run the query independently, ideally by a second analyst, and reconcile counts exactly.
- Check against another source, such as bed-management logs or the department register, for a sample of days.
- Look at the denominator. A rise in visits might reflect a new clinic opening or a change in coding. State what the hospital's total volume did over the same period.
- Compare against public data such as a state or national series, with a link, so that readers can see whether the hospital's pattern is local or general.
- Obtain clinical review of every sentence that interprets the data.
6. Write the story and the methods box
A good data release has a headline finding, one comparison, one chart and a clinician comment that explains what the finding means for the public. The methods box says what was counted, over what period, what was excluded, and what the numbers cannot show. Admit limits plainly: a single hospital's admissions are not the city's disease burden, and the story should not imply they are.
The clinician comment should follow the public-health education rule in clause 5.1 and clause 8.2(iii) of the NMC guidelines: it informs, does not promote the practice, and does not name the hospital's service as the solution. A line telling readers when to seek care, cleared by the clinician, is public information. A line inviting them to book a package is not.
A hypothetical worked example
A 300-bed hospital in Nagpur wants to tell a story about summer heat. Its analyst extracts weekly counts of emergency visits grouped under a clinician-agreed "heat-related illness" category, banded by age, for the last three seasons, with city of residence reduced to "Nagpur district" or "outside district".
The initial table has one cell, children under five in the first week of April, with three patients. That is below the floor of 10. The analyst merges the under-five and 5-to-12 bands for that week. Totals by week stay above the floor. The independent re-run matches exactly. The denominator check shows that overall emergency volume was flat, so a rise in the category is not just a busier department.
The story: "Heat-related emergency visits at one Nagpur hospital arrived four weeks earlier this season than the previous two." The numbers, the method box and the clinician advice on recognising heat illness go to the journalist. Nothing in the table could identify anyone, and the clinician's quotation is about prevention. All figures here are invented to show the mechanics.
