AI Medication-Safety App for Oncology

Every week, oncology teams are expected to check medication lists of patients who may be taking ten or more drugs at once. A manual check can take up to 45 minutes and still leave space for missed interactions, especially when genetics or trial eligibility are involved. Mistakes lead to hospitalizations, wasted treatment cycles, and financial losses.

In 2024, our team developed an AI app for oncology clinics that allows staff to:

• Instantly check all medications for potential conflicts and PGx risks
• Identify open clinical trials for each patient before it’s too late
• Cut review times and stay fully HIPAA-compliant, with easy integration into the EHR

In pilots, automated medication monitoring and trial matching with Oncogarde (see case) helped cut hospitalizations by half and reduced screening time from 40 to 12 minutes per patient. The clinic also saw fewer claim denials and a clear rise in trial enrollments.

This article details what we built, why it works, and how our approach differs from generic “AI in healthcare” platforms — with workflow-level examples, outcome metrics, and no speculation.

‍

How AI is Transforming Clinical Practice and Trials

AI is already tackling challenges in oncology clinics that were previously time-consuming and risky for patients.

Medication safety:

At Ballad Health (USA), pharmacists use the MedAware system directly within their EHR. It automatically highlights prescriptions where combinations fall outside standard protocols or carry hidden risks. In the first six months, the number of serious prescribing errors dropped by 53%. Pharmacists now spend less time on routine checks and more on patients who are truly at risk.

Clinical trial matching:

At TriHealth Cancer Institute and similar centers, Tempus and ConcertAI scan electronic records to identify suitable clinical trials. The entire process for a patient now takes 10–12 minutes instead of 40, and the number of patients enrolled in active protocols has nearly tripled. Thanks to automation, doctors can discuss trial options with each patient before enrollment windows close.

Pharmacogenomics:

At NorthShore Health, the ActX system checks every new prescription against the patient’s genetic profile. Instant alerts have led to dozens of therapy plans being changed within the first weeks, preventing serious side effects and difficult decisions around drug choice.

What makes Oncogarde different:

Unlike generic platforms such as MedAware, Oncogarde is purpose-built for the complexities of oncology. It not only flags medication risks, but also takes into account each patient’s genetic profile, automates clinical trial matching, and generates clear reports for both clinicians and management. This gives oncology teams a real-time, 360-degree view of risks and opportunities for every patient — not just standard drug safety.

AI products in clinics pay for themselves before the next budget cycle when they’re integrated into daily routines and clinicians can see exactly how decisions are made. Where staff trust the system, workload decreases, therapy adjustments are made faster, and more patients get a real chance to join relevant clinical trials.

If you’re planning a similar solution, our app cost calculator lets you see a detailed estimate—including budget, timeline, tech stack, and core development team—in just a few minutes. Simply describe your requirements in your own words to get started.

‍

Why AI Models Are Especially Effective in Healthcare

Artificial intelligence produces more accurate and reliable results in medicine than in most other fields. This is due to the unique structure and requirements of healthcare data and processes.

• Highly structured data: Medical records, test results, prescriptions, and outcomes are all recorded in a unified format and updated daily.
• Clear cause-and-effect: There are transparent success metrics — a patient recovers or not, complications are reduced or not.
• Large, well-labeled datasets: Healthcare generates millions of precisely labeled cases (diagnosis, treatment, outcome), unlike many other industries.
• Limited number of “right” answers: Unlike marketing or sales, medicine follows established protocols, so the range of correct solutions is narrow and well-defined.
• Real-life consequences: Every recommendation or mistake immediately affects a person’s health and is recorded in clinic statistics.
• Strict validation: All new AI models must pass independent clinical trials and demonstrate real-world effectiveness before deployment.
• High motivation to implement and verify: Doctors and clinics are motivated by fast, measurable improvements — unnecessary features are rarely adopted.

AI in healthcare quickly adapts by learning from thousands of patient cases, and its results are tracked in the clinic’s own monthly metrics.

‍

Limitations of AI in Healthcare

Even with all the data and clinical protocols in place, digital tools in healthcare still face real barriers. One of the biggest issues is trust: many systems offer recommendations but can’t clearly explain their reasoning. For a doctor, a vague alert isn’t enough—if the system can’t show its logic in plain language, it risks being ignored or even switched off entirely.

Technical integration is another sticking point. Most hospital software is a patchwork of legacy systems, making it a challenge to embed new technology without major disruptions. Many promising projects have stalled because their tools simply didn’t “talk” to the rest of the IT stack.

Lastly, patient privacy remains a top concern. Any mistake in data handling can lead not just to regulatory fines, but to real harm for patients and reputational damage for the clinic. That’s why the most successful solutions today are those that focus on transparency—showing exactly how data is used, keeping audit trails, and always putting the user (and patient) in control.

‍

Challenges in Medication Safety and Oncology Trials

In oncology, the workload for doctors and pharmacists is intense:

• A single patient is often prescribed ten or more medications at once
• Manual review of all potential interactions and genetic risks can take up to 45 minutes per case
• With limited time, it’s easy to miss a dangerous conflict or overlook a relevant clinical trial

Additional complications include:

• Alert fatigue — important warnings get lost among a flood of less relevant notifications
• Strict regulatory requirements: every decision must be justified and documented
• Lack of trust in digital tools if they can’t explain their recommendations or fit into daily clinical routines

There have been cases where advanced solutions (like Watson for Oncology or certain generative AI prototypes) gave recommendations that weren’t practical — and sometimes even put patients at risk. Technology delivers real results only when it adapts to the true clinical context and supports, rather than replaces, the expertise of the medical team.

‍

Oncogarde AI Halves Oncology Medication Errors

When a US oncology clinic first approached us, they were clear about their biggest pain point: pharmacists were spending up to 45 minutes reviewing each complex patient’s medication list, yet critical risks were still being missed and opportunities for clinical trial enrollment were slipping through the cracks. They weren’t asking for “just another AI tool” — they needed a solution that would actually hold up in real-world clinical practice.

Early discoveries:

From our very first meeting with the clinic team, we realized that a single patient’s information was scattered across three different places: the EHR, a separate lab system, and an old Excel file for PGx data. Every time a treatment plan changed, some records wouldn’t update until a day later. Doctors complained that approving a new order cost not just time, but a lot of nerves—double-checking all meds, searching for the latest lab results, and dealing with data that simply “disappeared” overnight due to manual syncs.

Teamwork and tech choices:

• We spent a week shadowing pharmacists on duty, documenting every workflow bottleneck—especially in the evenings and on weekends when IT support wasn’t available. It turned out that 30% of medication risks surfaced at night.
• To handle EHR integration, we had to spin up a separate server, since the main system didn’t support automated nightly updates. After discussions with IT, we added a cron script and manual morning validation to catch missing records.
• To save staff from reviewing the entire patient history every time, we built a “change timeline”: now, pharmacists see only what’s changed in the last 24 hours, which saves at least 10 minutes per review.
• PGx integration was phased: some data still arrives as photos or PDFs. We designed a lightweight entry form for rapid manual input, now used for all new patients.
• We piloted an “Ask care team” feature: doctors often want to discuss an alert with colleagues, not just with pharmacy. We added a “Comment” button so the team could flag and discuss tricky cases directly in the system.
• Not every recommendation fit the clinic’s needs: for instance, some “suggested” trials were closed or not available locally. We added a “trial status” flag, so doctors can immediately see if a trial is actually enrolling.

Results and recommendations:

• With the new change timeline and smarter alerts, staff now close all reviews within a single shift, instead of letting cases pile up.
• 80% of manual review errors occurred at night—after automating nightly imports, missed risks dropped to nearly zero.
• For the first time in years, the clinic’s research department had clean, real-time stats on clinical trial enrollment—every patient added is logged and reported automatically.
• We wrote step-by-step guides for every clinical role, which halved onboarding time for new hires.

Advice to other teams:

• Don’t rely on automation alone—really dig into how your clinic operates at night, on weekends, and during high volume.
• Always leave room for manual data entry and team comments—not everything is ready for pure automation at launch.
• Take time to map out real-world workflows in your own team, rather than copying “best practices” from other hospitals.
• Assign clear responsibility for updating reference data—otherwise, some features will quietly stop working over time.

A product like this only works when developers and clinical staff are in constant contact—at every stage, not just for a demo. That’s how you build a tool that isn’t just impressive in theory, but actually gets used every day.

‍

Comparing Oncogarde with Market Leaders

When we received the clinic’s request, our first step was a classic discovery phase—not because it’s industry best practice, but because rushing into automation can just replicate someone else’s mistakes.

In those first weeks, we dug into live demos, spoke directly with real users, and observed how competitor solutions performed under real-life pressure—not in sales presentations, but in busy clinical shifts where every minute counts.

MedAware

Very effective for catching prescription errors, especially at high volume. But when we tested it in oncology workflows, the system operated too generically, missing the nuance of complex regimens and patient genetics. Users told us, “The alerts just come in a flood, but the really tricky cases still end up as manual work.”

FeelBetter

Works well for general medicine and older adult care, especially with polypharmacy. For oncology and particularly for trial matching, though, the platform’s features just weren’t deep enough—and any extra customization landed back on the clinic’s IT team.

Tempus

Great for major centers: powerful molecular analytics and close integration with research. However, onboarding takes months, and budgets are geared toward big hospital systems. Clinicians in smaller practices told us they needed more flexibility and faster deployment.

ConcertAI

A strong choice for population-level analytics and research. But when it comes to daily decision-making at the patient’s bedside, many of the tools feel too heavy and slow to implement, making it hard for staff to get immediate value.

All of these findings shaped our approach to Oncogarde’s architecture: we built it to be light and quick to deploy, work with existing EHRs, provide transparent explanations for every alert, and offer built-in trial matching. We chose a blend of rule-based and machine learning logic to maintain clarity instead of chasing trendy “black box” AI features.

As a result, our system is valued where speed, accuracy, and real-world customization matter more than an idealized slide deck. Oncogarde isn’t for every clinic—it’s for those who want a tool that adapts to real practice, not the other way around.