What is the study about?
The study discusses how artificial intelligence and IT solutions can make the revenue cycle of hospitals and clinics faster and more accurate. It starts from a familiar diagnosis: manual processes, different rules for each payer and systems that don't talk to each other cause errors, rework, payment denials and delayed collections [1].
The paper was published in March 2025 in The American Journal of Engineering and Technology (vol. 7, no. 3, pp. 141–162), open access under a Creative Commons BY 4.0 license [1]. The context is the US healthcare system: private and public payers, CPT and ICD coding, and HIPAA privacy rules.
The text is divided into literature review, method, AI automation, IT solutions, discussion, results, limitations and recommendations.
The six technology groups the paper analyzes
The paper organizes the solutions into six groups. The table shows what the study attributes to each one and how we read it for a Brazilian hospital billing health plan operators.
| Technology | What the paper describes | How it applies in a Brazilian hospital |
|---|---|---|
| Robotic process automation (RPA) | Bots that run repetitive tasks: eligibility checks, claim submission, payment posting | Eligibility and authorization code lookups on payer portals, downloading payment and denial statements, reconciling receipts. It's the most mature technology on the list |
| Machine learning (predictive analytics) | Models that learn from history and flag claims at risk of denial before submission | Predicting denials by payer, reason and item. It depends on denial history organized by reason, ideally with TISS denial codes |
| Natural language processing (NLP) | Reading clinical notes and medical records to suggest the correct coding | Cross-checking medical record, prescription and nursing checks against the claim: item used but not charged, item charged without a record, OPME without justification |
| Anomaly and fraud detection | Algorithms that identify duplicate, unbundled or above-standard charges | On the hospital side, the same capability finds duplicates or missing charges before the payer does |
| Cloud and EHR integration | Real-time data and claim status tracking | Integration between the electronic health record (EHR), management system and billing, so the claim starts out complete |
| Blockchain and smart contracts | Immutable transaction records and automatic payment under agreed rules | The paper itself acknowledges that adoption is still limited [1]. In Brazil, data exchange already follows the TISS standard [4] and payment rules are set in the hospital–payer contract |
What barriers does the study point out?
The barriers are the most useful part of the paper, because they apply to any hospital, in the US or in Brazil [1]:
- Implementation cost. Software, infrastructure and training weigh more on small and mid-size hospitals, which tend to adopt later and lose more revenue in the meantime.
- Data protection. AI needs large volumes of patient and financial data. In the US, the reference is HIPAA; in Brazil, the General Data Protection Law (Law 13,709/2018), which treats health data as sensitive data [3].
- Legacy system integration. Legacy billing systems don't always connect to new tools, and data gets stuck in silos.
- Team resistance. Fear of losing jobs and unfamiliarity with digital tools slow adoption. The paper recommends presenting AI as support for professionals, not a replacement.
- Uncertain return. Because the financial gain appears gradually, many institutions postpone the investment.
Are the study's numbers a valid benchmark?
Not as targets. The paper presents striking percentages but doesn't say which hospitals, which period or which study each number came from. The method is described as secondary data analysis (papers, government and payer reports, case studies), with no list of databases or sample size [1].
| What the paper claims | Where it shows up | What's missing to use it as a benchmark |
|---|---|---|
| Denials account for 10% to 15% of submitted claims | Introduction and review | The references cited for this number are reviews of AI in medicine in general |
| RPA reduces denials by 30% | Automation section | Sample, period and source study |
| Verification algorithms reduce rejections by 30% to 40%; RPA cuts claim review time by 50% | Results | Sample, period and source study |
| Automated charge capture increases net revenue by 20% to 25% | Results | Sample, period and baseline |
| Reimbursement predictions over 85% accurate; cash flow 25% more stable | Results | Definition of the KPI and the sample |
| Blockchain cuts fraudulent charges by 70% and transaction costs by 40% | Results | Source study; the text itself says adoption is limited |
Other points deserve attention when reading. The charts by specialty (denials over five years, efficiency by department) don't cite their data source. The paper's record shows acceptance (02/21/2024) before submission (01/25/2025), probably a typo. And the reference list, after the first 35 or so on AI in healthcare, includes many works on IoT, supply chains and energy, with no direct link to the revenue cycle.
None of this invalidates the paper's conceptual map. But to estimate the gain from an AI project at your hospital, the starting point must be your own baseline.
US denials and Brazilian claim denials: how to compare
The paper talks about "claim denial", a refusal to pay a claim. The Brazilian claim denial is broader: it can reject the whole claim or just some items, and it's measured in different ways. That's why US percentages don't compare directly with Brazilian ones.
The main public benchmark in Brazil is the Anahp Observatory. At Anahp member hospitals in 2025, initial denials were 15.93% of amounts receivable from payers, final denials were 1.72% of gross health plan revenue and the average collection period was 77.35 days [2]. Figures for every year are in Industry data.
| Term in the study (US) | Closest equivalent in Brazil |
|---|---|
| Claim denial | Claim denial (full or partial) or payment refusal |
| Days in accounts receivable (A/R) | Average collection period |
| Eligibility verification | Eligibility verification (verificação de elegibilidade) |
| Prior authorization / pre-certification | Prior authorization |
| Charge capture | Posting items to the claim |
| Medical coding (CPT, ICD) | TUSS and ICD coding |
| First pass resolution rate | Claim acceptance on first review |
| Net collection rate | Collection rate (índice de recebimento) |
| EHR | Electronic patient record (PEP) |
| HIPAA | LGPD |
US denial causes match Brazilian ones
The paper includes a Pareto chart of denial causes, without stating the database. The order, however, is familiar to any Brazilian billing specialist: incomplete documentation, coding errors, eligibility problems and missing authorization account for most; duplicates, uncovered items and late submission come next [1].
The practical lesson holds in both countries: most of the loss starts before the claim is submitted, at the front desk, at authorization and in documenting care. That's where automation and AI have the most to do. Details on each cause are in the Claim denials guide.
What AI needs to work in the hospital
AI learns from history. If the history is poor, the model learns the wrong things. Before buying a tool, the hospital needs four things:
- Structured denial reasons. Denials recorded by payer, reason and item, ideally with TISS codes, for at least 12 months.
- A measured baseline. Initial denials, final denials, average collection period and discharge-to-submission time, with a fixed formula. Without it, there's no way to know whether the tool worked.
- Integration between systems. Medical records, management system and billing talking to each other, so the tool sees what was done and what was charged.
- Data protection rules. A legal basis for processing, a vendor contract covering LGPD obligations and involvement of the hospital's data protection officer [3].
Where to start
- Organize your denial history by payer, reason and item for the last 12 months.
- Measure the baseline with the revenue cycle KPIs, always with the same formula.
- Pick one step with high volume and clear rules for the first project: eligibility and authorization, claim review before submission or reading payer statements.
- Compare before and after in the same period of the year, and, if possible, with a group of claims that didn't go through the tool.
- Involve the data protection officer and the team from the start: the LGPD and people's buy-in set the pace of the project.
Download the full study
Optimizing Revenue Cycle Management in Healthcare: AI and IT Solutions for Business Process Automation
Abdullah al Mamun, Muhammad Saqib Jalil, Mohammad Tonmoy Jubaear Mehedy, Maham Saeed, Esrat Zahan Snigdha, MD Nadil Khan and Nahid Khan · The American Journal of Engineering and Technology, vol. 7, no. 3, pp. 141–162, 2025 · PDF in English, 22 pages, 0.6 MB
Published under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. The PDF is distributed unchanged. The analysis above is Accrua's and was not reviewed by the authors.