Asistent vykazování AVD
A Czech hospital coding assistant that reads an anonymised patient discharge letter with machine-learning language models and proposes the ICD-10 diagnosis codes to report, launched straight from the hospital's clinical information system.
What is Asistent vykazování AVD?
Asistent vykazování AVD, styled AVⒹ on its own pages, is a diagnosis coding assistant for Czech hospitals published by ICZ.HEA a.s., part of the ICZ group. It addresses a narrow but expensive problem: turning clinical documentation into ICD-10 codes is slow, easy to get wrong, and directly determines what an insurer reimburses. The vendor presents it as the first coding support solution on the Czech market to use artificial intelligence.
The assistant reads the patient's discharge letter and returns the diagnoses that should be reported. It runs four machine-learning language models, two covering all diagnoses and two dedicated to identifying the principal diagnosis. According to the product sheet these were trained on coding data from Fakultní nemocnice Brno, a hospital chosen for the quality of its coding and the breadth of care it delivers.
Two tasks are offered. AVⒹ, poraď mi asks the assistant to suggest codes from scratch. AVⒹ, zkontroluj mi submits an already coded episode, and the assistant compares the two lists, marks the diagnoses it agrees with, flags those it finds ambiguous and adds any it believes were missed.
Output is returned at three-character and four-character precision, each diagnosis carrying a probability of correctness. Recommendations are surfaced above an individual threshold per diagnosis or above 40 percent. Each entry shows whether a more precise five-character code exists and the severity level on the current CZ-DRG scale of 0 to 4. Users can filter, open details, view the anonymised source letter and cross-check against PVD, the publisher's other coding product.
It is delivered as a service, the server side sitting either in ICZ's data centre or at the customer, and reached through a browser from a button inside the hospital's clinical information system. One limit is stated plainly: there is no write-back, so approved codes are copied to the clipboard and pasted in by hand.
What it does
- Propose the diagnosis codes to report from an anonymised discharge letter
- Check diagnoses a coder has already entered and flag what is missing or ambiguous
- Attach a model-computed probability to every diagnosis returned
- Single out one or more candidates for the principal diagnosis
- Show the CZ-DRG severity level and whether a more precise code exists
- Filter results by severity or probability and open a diagnosis in detail
- Copy selected codes to the clipboard for entry into the clinical system
When to use Asistent vykazování AVD / When not to
A quick filter to help you decide if Asistent vykazování AVD is the right fit.
When to use Asistent vykazování AVD
- Hospital medical coders responsible for reporting inpatient care to health insurers
- Heads of DRG and coding departments who need cover when a coder stands in for another specialty
- Trainers onboarding new coders, who can work case by case against the assistant's suggestions
- Coding auditors who need to re-check a large number of already coded hospital episodes quickly
- Hospital controlling and revenue teams tracking under-coding and reimbursement shortfalls
When not to use Asistent vykazování AVD
- Healthcare providers outside the Czech Republic: the tool is built on the Czech ICD-10 modification, the CZ-DRG severity scale and Czech insurer reporting
- Organisations wanting a self-service product: there is no sign-up, no published price and no advertised trial for this assistant
- Hospitals unwilling to run an integration project, since the clinical information system must anonymise the discharge letter and call the assistant itself
- Teams expecting codes to be written back automatically into the patient record, which the vendor states is not supported
- Clinicians looking for diagnostic decision support: the tool codes documentation about care already delivered, it does not advise on treatment
How to use Asistent vykazování AVD
A typical end-to-end flow, from setup to results.
- Request a demonstration by writing to the product mailbox, which is the only advertised route to the tool
- Agree an integration project with the publisher, which installs and administers the dedicated server running Debian 12 or later
- Have the hospital's clinical information system anonymise the discharge letter before transmission
- Have that system collect the accompanying data: patient age and sex, discharging ward, and the already coded diagnoses for check mode
- Connect the clinical system to the assistant through its web services
- Add the two task buttons at a suitable place in the clinical system's screens
- On a given inpatient episode, click either the advise task or the check task
- Let the clinical system pass the data and open the assistant's portal, already populated with the results
- Filter the returned diagnoses by severity or probability and open individual entries for detail
- Copy the retained codes to the clipboard and paste them into the clinical system to complete the report
Pros & Cons
Pros
- Substantial time saved on routine episodes, with results returned in a few seconds
- Surfaces codes a coder had not noticed and flags those that raise the recorded severity of a case
- Trained on the coding of a hospital selected for its quality, which the vendor credits for accuracy and error reduction
- Makes it easier for a coder to stand in for a colleague from another specialty
- Useful for onboarding new coders and as a fast instrument for coding audits
- Reachable from the clinical system through a browser, with nothing to install on workstations
- Unusually precise functional documentation, including the threshold, the severity scale and the named training corpus
Cons
- No write-back into the clinical system, so approved codes must be copied and pasted by hand
- Requires an integration project on the hospital side, with immediate compatibility limited to five clinical systems
- No published price, no rate card and no self-service trial: everything goes through an email request
- No privacy policy, no terms and conditions and no legal notice on the site, verified against the full 36-page inventory
- No data processing agreement, no subprocessor list and no security certification, for a product handling inpatient documentation
- No accuracy or recall figures and no independent evaluation, with models trained on a single hospital's data
- Tied to the Czech market by design, through the Czech ICD-10 modification, CZ-DRG and Czech insurer reporting
Pricing & Plans
No price is published. There is no pricing page anywhere on the site, the usual Czech pricing paths return errors, and no amount in any currency appears on the collected pages or in the product sheet. The stated route to the product is a demonstration request sent by email to the publisher, after which the tool is delivered as a service with the server component hosted either in ICZ's data centre or at the customer. Readers should note that the time-limited free access advertised on this site belongs to PVD, a different product from the same publisher, and does not apply to this assistant.
Data, GDPR & hosting
A consolidated view of how Asistent vykazování AVD handles your data.
GDPR overview
GDPR implementation is partial and unevenly documented. Concrete elements exist: the publisher ICZ.HEA a.s. is named as controller, a data protection officer is designated by name (Jan Mašek, jan.masek@i.cz), consent is collected at registration and can be withdrawn by email, the cookie policy of 15 January 2025 enumerates the rights of access, rectification, erasure, withdrawal, portability and objection, and it names the Czech supervisory authority. Against that, the material gaps are substantial for a product handling inpatient medical documentation: the site publishes no privacy policy, no terms and conditions, no legal notice, no data processing agreement and no list of subprocessors. The contact block closing the cookie policy also carries a corrupted address left over from the template.
Who owns the data?
The site publishes no product privacy policy and no terms and conditions, so data ownership for the assistant is never stated. The only named controller is the publisher, ICZ.HEA a.s. (company number 07240091), on the site's GDPR notice, which designates Jan Mašek as data protection officer. On the product side the single written rule is that the hospital's own clinical information system anonymises the discharge letter before anything is sent. The product sheet adds that the server component sits either in ICZ's data centre or on the customer's premises. Who owns the processed documentation, who may reuse it and for how long is left undocumented.
Reuse rights
No terms of use exist, so nothing authorises or restricts what a customer may do with the output. In practice the assistant returns a list of candidate diagnoses that the coder copies to the Windows clipboard and pastes into the clinical system, which is the reuse path the vendor describes. Inputs are the anonymised discharge letter, the patient's age and sex, the discharging ward, the already coded diagnoses in check mode, hospital data and user identification, the last of these stated to serve usefulness evaluation. On training, the product sheet discloses that the four models were learned on coding data from Fakultní nemocnice Brno; it takes no position on whether customer data feeds later training, and no opt-out is documented.
Data retention & training
Hosting summary
The publisher describes where the software runs but never names a country. The product sheet states that the assistant is delivered as a service with the server component placed either in ICZ's data centre or at the customer, and that it can be operated within the ICZ cloud, a hospital's own cloud or that of a hospital group. The technical page adds that it is a web application on a dedicated server running Debian 12 or later, with installation and server administration handled by ICZ.HEA a.s. Users reach it through an ordinary browser, so nothing is stored on workstations. No hosting jurisdiction, region or subprocessor is disclosed. One point should not be confused with the above: the marketing website itself resolves to an address in Slovakia, which says nothing about where the product processes hospital data.
Where Asistent vykazování AVD works
Country-level availability.
Available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Asistent vykazování AVD.
- Output is probabilistic, surfaced above a 40 percent threshold or an individual one per diagnosis, so a coder who accepts suggestions without reading the record will propagate model error into insurance claims
- No accuracy or recall figures and no independent evaluation are published, and the models were trained on a single hospital's coding, so performance elsewhere is undocumented
- Because the tool speeds up routine coding, sustained use risks eroding the manual skill it depends on, which matters most for the newly trained coders the vendor recommends it for
- Anonymisation of the discharge letter is performed by the hospital's own system, so responsibility for de-identifying patient data sits with the customer, not the publisher
- No privacy policy, terms, legal notice, data processing agreement or subprocessor list is published, and the cookie policy's contact block contains a corrupted address left over from a template
- The published details are inconsistent: the registered office postcode differs between pages, and the downloadable product sheet still names the former entity ICZ a.s. and a group marketing address rather than the current publisher
- The website itself is hosted in Slovakia while the publisher and market are Czech, and the hosting location of the product is never stated
Setup & Integrations
Technical difficulty
High, and it is an integration project rather than a sign-up. The hospital must do three things in its own clinical information system: anonymise the discharge letter and gather the accompanying data, call the assistant through its web services, and place the two task buttons where coders will use them. The publisher installs and administers the dedicated server. The vendor states that everything needed is already prepared on the assistant's side, and five clinical systems are listed as immediately integrable, which cuts the effort sharply for those hospitals. Nothing is installed on workstations.
Deployment
Integrations
Supported languages
Behind Asistent vykazování AVD
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Asistent vykazování AVD actually do?
What are the two tasks it offers?
What were the models trained on?
Does it write the codes back into the patient record?
Which clinical information systems are supported?
Where does the application run?
How much does it cost?
Is there a free trial?
Is there an API?
Who publishes the tool?
Should you pick Asistent vykazování AVD?
Asistent vykazování AVD is a narrow, seriously built product whose documentation is better than its commercial packaging. On the functional side the publisher is unusually forthcoming: it names the number of models, the hospital whose coding data trained them, the recommendation threshold, the severity scale used, the exact inputs collected and, notably, a limitation that most vendors would leave unsaid, namely that codes cannot be written back into the clinical system and have to be pasted in by hand.
The audience is correspondingly narrow. This is a tool for coding and DRG departments in Czech hospitals, bound to the Czech ICD-10 modification, the CZ-DRG severity scale and reporting to Czech health insurers. Within that audience the case is credible, particularly for coding audits, for onboarding new coders and for covering specialties a coder does not usually handle.
The weak side is everything around the software. There is no published price, no rate card and no self-service entry point. More seriously for a product that processes inpatient documentation, the site carries no privacy policy, no terms and conditions, no legal notice, no data processing agreement and no subprocessor list, and no security certification is displayed. No accuracy figures or independent evaluation are published either, and the models come from a single hospital's data, so how well they generalise is undocumented.
A Czech hospital evaluating this assistant has enough to judge what it does and will need a contract discussion to establish everything else. The probabilistic output makes human review indispensable rather than optional, which the vendor's own framing already implies.
- Choosing a selection results in a full page refresh.
- Opens in a new window.