“It was a rainy day in Palo Alto. The line in front of the Apple Store was interminable, and only a few strong-willed people could be persistent enough to wait for hours and have a shot at earning one of the first commercially available Apple Vision Pro. Each step brought me closer to a revolution in computing…”

Ok, let’s be real. This sounds great as an opening, but I would never wait in line for any commercial product. However, thanks to the amazing Digital Health team of the Stanford Byers Center for Biodesign, I was able to try the new Apple Vision Pro and have some discussion about its potential in computer vision and healthcare.

Biodesign for Digital Health

Why Computer Vision and Healthcare

I won’t talk about the manufacturing quality. Apple usually does an excellent job and their new top-of-the-line product does not underdeliver. There are already a lot of professional reviews talking about this, not really my field. Let’s look at things from a computer vision and AI perspective — what is the potential, what should we expect, and what are the future lines of research?
And, of course, being a surgeon, it’s difficult not to think about the subsequent developments of this type of technology in healthcare. My head bumped into enough screens in the operating room to realize that they might not be the best solution for our current applications. Screens are everywhere in healthcare: endoscopic, laparoscopic, exoscopic, robotic surgery, patient monitoring (ECG, saturation, ventilation), radiology, etc. And if you think the Apple Vision Pro is expensive, you should look at the prices of medical-grade monitors.
But screens were a necessary evil that allowed us to shift from purely optical visualization technologies (e.g., optical microscopes and loupes) to a direct digital input — the dream of every computer vision researcher. With this type of input, it’s possible to collect training data (data that we previously threw away!), analyze procedures, and develop AI applications that are valuable in clinical practice.

From Eyes to Algorithms

While my initial focus was testing visual quality, I was struck by how intuitive the experience was, and I found myself being more impressed by some less-discussed features:

- Eye-tracking

- Passthrough quality

- Hand/pinch detection and tracking

These are the elements that make the experience particularly seamless and that, together with the concept of “Spatial Computing,” will revolutionize current UI/UX design standards. But they also made me more aware of the possible interactions with computer vision and AI.

Let me explain.

Eye-tracking is not just the “new mouse”; it’s a data source

Attention is a central concept in AI and computer vision. The article “Attention is All You Need” introduced this concept by describing the transformer architecture in natural language processing, and “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale” transferred this concept to the visual field. We are talking about a mathematical form of attention, but the broad meaning is still there.

Eye movements are the external manifestation of human visual attention. For the first time, we (well, Apple) will have an instrument that tracks and records eye movements prospectively, long term. This is an exciting concept since eye micromovements are mostly involuntary and will help us understand how we perceive the visual world — with potential correlates on the way we structure and evaluate computer vision algorithms. On the other hand, it’s scary to consider that, by virtue of their subconscious nature, eye movements might help profile customers or track what the person is thinking — a sort of “you are what you look at” concept.

Finally, medical conditions (especially neurologic and balance disorders) can influence eye micromovements and eye-hand coordination. Constant tracking may be a beneficial “opportunistic screening” tool to diagnose these conditions early.

High passthrough quality brings computer vision to vision

One of the most striking elements is the impact of high passthrough quality on the overall experience. The objective is to shut off the perception of looking at a screen and to integrate digital elements into the real world. And the Apple Vision Pro gets really close to that.

This has been achieved thanks to the concomitant increase in resolution and quality of camera sensors and micro-OLED displays — and we are getting closer to a condition where it will be impossible to determine if we are looking through a digital camera + screen, or through a glass.

As a consequence, it will be possible to apply computer vision to every setting in everyday life. It’s not just autonomous driving and specific applications. Computer vision applications won’t need a separate device — smartphones, tablets, computers, endoscopes — the interaction will be direct.

Spatial computing is the perfect platform for computer vision.

Interfaces based on hand-eye input will change UI/UX design principles

Conventional interfaces are based on well-defined input devices (e.g., mouse, keyboard, trackpad). Here, everything in the visual field can potentially become an input source — starting from the hands and extending to the entire available space. Again, this is based on computer vision (eye and hand/gesture tracking on top of everything) — the entire video feed from the numerous cameras must be processed as an “input”, reprocessed, and integrated with digital components (creating the “output”), shattering the conventional separation between input and output. This will significantly increase the interactions between the applications, user, and environment — ultimately requiring new UI/UX design paradigms.

Yes, And What About Healthcare?

As a physician and surgeon, it’s difficult not to think about the potential revolution this technology would bring into healthcare, apart from the previously mentioned struggle between my head and floating monitors. The surgeon could easily position and look at 2D or 3D screens during endoscopic or exoscopic surgery, integrating the view with the patient’s information from vital signs tracking, radiologic imaging, and image enhancement techniques. The view could be further extended with dedicated computer vision algorithms to detect instruments, tissues, and anatomical structures.

Finally, UI/UX design in healthcare is far from ideal. Ease of use and functional layouts are often low priorities when dealing with complex medical data. However, the advent of spatial computing offers a blank slate to build on, maybe following better concepts of design and usability.

The Spezi framework from Stanford caters to these needs thanks to its modular structure. Specifically, Spezi is an open-source framework for the rapid development of modern, interoperable digital health applications, and the team is already working on integrating applications in VisionOS.

Spezi

In wrapping up my dive into the Apple Vision Pro and its intersection with computer vision and healthcare, it’s clear we’re on the cusp of a transformative period. This device isn’t just about sharper images or smoother interfaces; it’s about redefining our interaction with technology and its application in medicine. The Vision Pro exemplifies how technology can seamlessly integrate into our lives, offering insights that extend far beyond the screen.

My Experience with the Apple Vision Pro and Future Perspectives in Computer Vision and Healthcare was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

The current problem in scientific research

We live in an era dominated by technology and AI, yet our access to scientific literature remains substantially unchanged in the last 50 years.

Let’s take medical literature as an example. PubMed remains the main access point for most researchers, and it functions as an old-style repository. There are no data aggregation functionalities; the search is based on basic criteria that mostly only rely on the article’s title, abstract, and keyword. There are no ways to interact with or analyze the content. No advanced data visualization.

And it isn’t PubMed’s fault. It does what it is intended to do: give access to research from a wide variety of Journals.

However, the current research process is clunky and time-consuming. As researchers and medical practitioners, when we need to get updated information on a specific topic, the process is usually the same (either with PubMed or other scientific repositories):

• Check for systematic reviews and meta-analyses.
• Aggregate and compare the results of the different systematic reviews and meta-analyses.
• Read and analyze the main articles in the reviews.
• Check for cross-references.

If reviews are not available:

• Check for recent articles on the topic.
• Get an idea of the overall trend.
• Compare patient numerosity in the different articles, level of evidence, and research-group experience.

Formulating a quick and impartial perspective on a specific subject is often difficult. One typical issue is anchoring bias, which occurs when individuals disproportionately depend on the initial information they receive about a topic. Additionally, selection bias plays a big role when the available literature or data is chosen based on accessibility or the author’s preferences rather than a comprehensive overview of the subject. These are just a few examples of the various issues that can skew judgment in forming opinions during the research process.

How can AI help?

Dedicated solutions

Some dedicated solutions for literature search and review are currently available. This is a great sign, and we are moving in the right direction. However, their practical utility in specialistic fields is still limited. Some examples are:

Connected Papers (www.connectedpapers.com), Litmaps (www.litmaps.com), and Research Rabbit (www.researchrabbit.ai) visually represent the research network on a topic connecting relevant articles. The outcome is a graphical literature map that can provide an overall view of the articles around that domain. Other than visual representation, these tools are also helpful in identifying cross-references.

Elicit (www.elicit.org) aims to use AI to answer research questions by summarizing the available literature. However, the current outcome is still far from a comprehensive analysis of all the articles. It can be useful for quick data extraction, but it sometimes misses key publications, and it’s not possible to manually upload a series of papers we are interested in. It’s an intriguing approach, but to date, its ideal target is the non-expert who wants to get scientifically-proven generic insights on a new topic.

Finally, Scite (www.scite.ai) allows users to see how a scientific paper has been cited by providing the context of the citation and classifying it as supporting, mentioning, or disputing the cited claim. This is an excellent addition to the toolbox, but the aim is to assist the experienced researcher in the conventional review process rather than creating an entirely new path.

Generalized approaches

Other than dedicated solutions, conventional LLM-based AI tools can also be beneficial.
Difficult not to mention ChatGPT — difficult not to mention it anywhere in the AI field, to be honest. However, its use cases are limited to generic topics and superficial knowledge that do not require high-level expertise. The web search tool can help in getting up-to-date knowledge, but the search is not limited to the academic literature and is frequently too generic. Hallucinations are also a significant problem; they can be pop up in the main text (quite rarely) or in citations (pretty frequently).
Well, let’s avoid making this error:

Humiliated lawyers fined $5,000 for submitting ChatGPT hallucinations in court: 'I heard about this new site, which I falsely assumed was, like, a super search engine'

Perplexity.ai (www.perplexity.ai) is another interesting option. It is possible to limit the search to the academic domain, and I’ve found that in terms of information retrieval and summarization, it provides better results than ChatGPT — at least in the medical field. Nevertheless, it’s far from summarizing specialized topics comprehensively since it tends to focus on the first search results.

A personalized pipeline for using LLMs in literature reviews

A potential option is to develop simple research pipelines based on basic Python code and LLM APIs to extract and summarize the needed data. There are different approaches, and this is a quick practical example.

Step 1 — Preliminary screening

In highly specialistic topics, we want to be sure that the articles we select for the initial screening are comprehensive. Unfortunately — for now — this is still a manual step (at least in my hands). However, there are some tricks. For example, ChatGPT helps convert a narrative research question into an effective PubMed research string. This might require some iteration, but it saves a lot of time down the line. You also might want to perform different searches with distinct strings.

Step 2 — Exporting results

After selecting the articles, you can press “Send to” on the PubMed research page and create the file.

Step 3 — Importing results in a citation manager

If you do research consistently, you probably already use a citation manager. My example is with Zotero (www.zotero.org), but it’s easily applicable to all the other apps. Here, you can merge and rearrange articles, exclude duplicates, and have an overall view of the collection.
With Zotero, it is possible to export a library in CSV format containing all the relevant information (including the full abstract), allowing you to subsequently manage it as a Pandas Dataframe.

Step 4 — Processing the data

First, let’s import the dataset:

import pandas as pd
df = pd.read_csv('Review_2024.csv')
df.head()

In this case, the main fields I’m interested in are: ‘Title’ and ‘Abstract Note’ (column containing the full abstract). Many other fields are available for different types of analysis.

At this point, it’s possible to automatically analyze and categorize titles and abstracts using an LLM. I think that Gemini Pro is a good choice at this time. Two main reasons:

• The outputs are of adequate quality
• The API is FREE! (For now)

This is just a quick example:

import google.generativeai as genai
import json

GOOGLE_API_KEY = 'YOUR_API_KEY'
genai.configure(api_key=GOOGLE_API_KEY)

def query_gemini(title, abstract):
    model = genai.GenerativeModel('gemini-pro')

    prompt = ("""
Given the following title and abstract, output a text in JSON format specifying:
- "endoscopy" : "yes" or "no" - If related to endoscopy or not
- "subspeciality": "Laryngology", "Head and Neck Surgery", "Otology", etc
- "number_of_patients_analyzed": "number of patients included in the study"
- "objective":"study objective here"
- "objective_short": "very short summary of the study objective in max 5 words"
- "AI_technique" : "Resnet", "Generic CNN", etc
- "Short_summary" : "Short summary of the study"

              """
              "\n\nTitle: {}\nAbstract: {}\n").format(title, abstract)

    response = model.generate_content(prompt)

    return response.text.strip().lstrip('``` JSON\n').rstrip('\n```')

Here, we structure the prompt using the JSON format, and then we get it ready to be added to the dataframe. After different tries, this is the way to make the output more consistent and easier to parse.

columns = ['endoscopy', 'subspeciality', 'number_of_patients_analyzed', 'objective', 'objective_short', 'AI_technique', 'Short_summary']
for col in columns:
    df[col] = None

Then, let’s add the required columns to the main dataframe.

for start in range(0, len(df), 300):
    end = start + 300
    responses = []

    for index, row in df.iloc[start:end].iterrows():
        response = query_gemini(row['Title'], row['Abstract Note'])
        try:
            parsed_response = json.loads(response)
            for col in columns:
                df.loc[index, col] = parsed_response.get(col, None)
        except json.JSONDecodeError:
            print(f"Failed to parse response for row {index}")
    chunk_df = df.iloc[start:end]
    chunk_df.to_excel(f'/YOUR_FOLDER/dataset_chunk_{start//300 + 1}.xlsx', index=False)

Finally, let’s start the process. I prefer to create different dataset chunks (every 300 articles, in this case) during the process. Gemini Pro is free until 60 QPM but it sometimes tends to limit access for a few minutes after a high number of requests. With this approach, you’ll be able to retain most of the work even after a crash.

After all of that, you’ll have a fully categorized literature search. It’s possible to analyze the distribution of topics (either with conventional statistical software or ChatGPT), get an overall idea of the different articles thanks to short summaries, etc. This is an interesting approach when you are thinking about writing a systematic review and meta-analysis following the current research and reporting guidelines, such as the PRISMA statement.

Future perspectives

The process is still time-consuming and highly manual. A potential development would be the introduction of “smart” article repositories that would automatically execute these parsing, classification, and summarization steps as an article is added. This automation would streamline the research pipeline and enhance accuracy and consistency in processing and presenting information. Let’s face it, systematic reviews and meta-analyses should not be performed by groups of researchers at a single point in time but should be interactive and updated after each article is added to the topic. This type of scientific dashboard would significantly enhance how research is accessed and distributed.

How To Use AI To Improve the Literature Review Process was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

The performance of large language models (LLMs) is growing at a breakneck pace. Models provide more coherent and consistent answers, with a progressive and significant reduction of hallucinations. This improvement is mainly related to the overall optimization of model architecture and training data, as well as the constant increase in parameters. Nonetheless, hallucinations are still occurring, sometimes in unexpected ways, and it is still a significant challenge to trace the source of these anomalies. Our understanding of the inner workings of LLMs is less detailed than what one could assume when looking at their widespread application, and this is a considerable limit in specific domains where random and unexpected errors could lead to severe consequences (e.g., healthcare and finance).

Neurodevelopmental Correlates in Artificial Intelligence

Understanding the neural development process in humans can be a useful guide to design and optimize LLMs. The human brain, especially during its development, undergoes various processes that enhance its efficiency and functionality, ensuring that neural circuits are adapted based on environmental interactions. During fetal development, the brain undergoes rapid growth, with an overproduction of synaptic connections. Subsequently, as the individual matures, synaptic pruning refines this neural network by removing redundant connections, thereby enhancing the groundedness and efficiency of neural processes. Neuronal activity plays a key role in this process - synapses that are frequently used and activated are strengthened and preserved, while those that are seldom used are pruned away.

Evolution has selected for this approach in network formation: construction through overabundance and subsequent pruning.

These neural networks are much more robust and efficient than networks that are constructed through other means (1).

Neuropathology of Biological Neural Networks

From the pathological viewpoint, alterations in synaptic pruning are one of the proposed etiological mechanisms behind neurological and psychiatric disorders (2). On one side, over-pruning can contribute to the excessive loss of functional synapses, such as in Alzheimer’s disease. On the other, unbalanced pruning is one of the potential causes of disorders such as autism and schizophrenia, where an inefficient “fine-tuning” of synaptic connections might lead to characteristic disease presentations. A dysregulated pruning process may be the source of symptoms such as hallucinations, delusions, and disorganized thinking in schizophrenia, as well as sensory processing challenges and behavioral profiles typical of autism.

Neuropathology of LLMs

In LLMs, ‘hallucinations’ typically refer to the generation of nonsensical or nonfactual content that is not grounded in the source material. However, Smith et al. (3) proposed the term ‘confabulation’ as a more fitting descriptor. In contrast to hallucination, confabulation involves generating incorrect narrative details, influenced by the model’s existing knowledge and contextual understanding, rather than implying any form of sensory perception. This redefinition aligns more closely with the operation of LLMs, which synthesize outputs based on patterns learned from vast datasets rather than experiencing sensory inputs.

In general, the extensive training of LLMs mirrors the initial stages of brain development, where a vast array of neural connections are formed. However, like the human brain, LLMs may require an adjunctive refinement process. This refinement, analogous to synaptic pruning in human development, would involve cleaning and optimizing the model’s architecture. Without this process, an LLM may risk being overwhelmed by ‘white noise’ — excess information or connections that obscure or distort the intended output.

So, the continual improvement of LLMs through methods like pruning may be required to ensure that their outputs are relevant, accurate, and grounded in the source material. As discussed above, these characteristics are particularly relevant when operating in fields where the reliability of the information provided by the LLM is critical for safety reasons. In fact, Elaraby et al. (4) showed that, to date, LLM-generated summaries in the legal or health space sometimes contain inaccurate information with the potential for a real-life negative impact.

A Technical Perspective on LLM Pruning — “Optimal Brain Damage”

As already described in technical papers, LLM pruning involves the reduction of model size by eliminating weights that contribute minimally to model performance, thus generating a sparser model. This process leads to more efficient models that require less computational power and resources while maintaining or even enhancing performance.

Pruning in LLMs is strikingly similar to synaptic pruning in neurodevelopment. Just as synaptic pruning optimizes neural pathways by removing redundant connections, model pruning in LLMs aims to maintain or enhance performance by removing redundant weights​​.

A fascinating description of the potential impact of model pruning was provided as early as 1989 by LeCun et al. (5) in a paper titled “Optimal Brain Damage.” As stated by the authors, by removing unimportant weights from a network, several improvements can be expected: better generalization, fewer required training examples, and improved speed of learning and classification. However, rather than brain damage, this can be viewed as a physiological step in neural structure optimization, a tailored “psychosurgical” approach aimed at favoring the adequate “maturation” of the architecture.

Indeed, it is extremely interesting to notice that, as recently demonstrated by Chrysostomou et al. (6), pruned LLMs tend to hallucinate less than their full-sized counterparts, potentially due to a greater reliance on source input rather than parametric knowledge from pre-training​​.

The Bigger, the Better?

The absence of adequate pruning might be one of the components underlying the presence of hallucinations in LLMs, and improvements in this field may lead to better models without the need for a significant increase in size, challenging the “the bigger, the better” stereotype. However, like synaptic pruning, AI model pruning is a balancing act of removing excess while preserving essential functionalities. The convergence of these biological and computational processes shows a parallel in seeking optimized efficiency and functionality in complex systems.

References

1. Navlakha S, Barth AL, Bar-Joseph Z. Decreasing-Rate Pruning Optimizes the Construction of Efficient and Robust Distributed Networks. Graham LJ, ed. PLoS Comput Biol. 2015;11(7):e1004347. doi:10.1371/journal.pcbi.1004347

2. Xie C, Xiang S, Shen C, et al. A shared neural basis underlying psychiatric comorbidity. Nat Med. 2023;29(5):1232–1242. doi:10.1038/s41591–023–02317–4

3. Smith AL, Greaves F, Panch T. Hallucination or Confabulation? Neuroanatomy as Metaphor in Large Language Models. Berkovsky S, ed. PLOS Digit Health. 2023;2(11):e0000388. doi:10.1371/journal.pdig.0000388

4. Elaraby M, Zhong Y, Litman D. Towards Argument-Aware Abstractive Summarization of Long Legal Opinions with Summary Reranking.

5. LeCun Y, Denker JS, Solla SA. Optimal Brain Damage.

6. Chrysostomou G, Zhao Z, Williams M, Aletras N. Lighter, yet More Faithful: Investigating Hallucinations in Pruned Large Language Models for Abstractive Summarization. Published online November 15, 2023. Accessed November 27, 2023. http://arxiv.org/abs/2311.09335

Psychopathology of Large Language Models: Foundation Models in a Neurobiological Perspective was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

Being an oncologic surgeon is my primary job and passion. It allows me to interact with people and immerse myself in the healthcare system, not the fancy corporate Healthcare, just everyday medicine.
And, as a researcher in AI, I’m noticing a growing disconnect between the actual clinical practice and the prevailing objectives of AI researchers and companies. This is, of course, just a personal opinion and not a critique of the current R&D processes, but it is a reflection grounded on some experience in both fields.

The disruptive potential of AI in customer software and industry is now clear. However, we must acknowledge that AI in healthcare is an entirely different animal; the degree of complexity, regulation, and risk is significantly higher than that of most other applications. Also, publicly available datasets are orders of magnitude scarcer than in many other domains due to privacy and accessibility limits.

So, big blockers and a higher level of complexity.

I’m currently staying in the Silicon Valley as a surgeon with a technical background in AI, which gave me direct access to this vibrant “ecosystem.” Meetings and conferences on AI are the order of the day. However, it’s difficult not to notice some facts:

• Clinicians do not participate in AI events.
• Clinicians do not participate even in AI for Healthcare events.
• The AI healthcare research is driven by the technical side, with minimal feedback/collaboration from clinicians.
• Even among clinicians, there is insufficient collaboration regarding data sharing and technical development.

A Tech-guided Approach

Firstly, the enthusiasm towards new technologies pushes us to try to apply them to every problem: “If the only tool you have is a hammer, you tend to see every problem as a nail,” in the words of Abraham Maslow. And I absolutely understand this tendency. AI is our new Thor’s hammer; why wouldn’t we want to try it on anything even remotely appropriate?

However, this directs research and progress focused on solving “technical puzzles” without answering a fundamental question. On one side, we can find amusing representations of this concept, such as the “That’s what she said” joke identifier (an amusing solution, I’m not criticizing); and, on the other, examples where the forced implementation of complex deep learning workflows is expensive and unnecessary.

Secondly, typical “top-down” strategies are based on market analysis and market-share calculation. In brief, “Let’s find a big and profitable field in healthcare, and let’s jam-pack it with AI.” As always, it might be a great short-term strategy, but the magic disappears after a while.

These approaches are rarely effective in healthcare. Physicians and surgeons often revert to conventional practices when the advantages of the new solution are not evident. Planck’s principle can be safely applied to medical innovation, “science advances one funeral at a time.” For this reason, a 5–10% increase in operational efficiency, while significant at scale, is hardly applied in the medical setting— we need a 2x-10x improvement in areas relevant to everyday clinical practice.

From “bench” to bedside

A practical approach would be to identify an actual problem, assess the efficacy of current solutions, and evaluate if AI can be employed to develop better solutions — the typical Mom Test.

How to talk to customers & learn if your business is a good idea when everyone is lying to you.

Currently, most major developments in AI for Healthcare are coming from Tech research groups and Tech companies. This association explains why the focus is skewed more towards the computer science side than the healthcare component.

In order to solve this issue, the direct involvement of clinicians and surgeons will be essential.

The need is clearly there. Medicine remains a highly flawed and subjective matter; calling it an “art” is looking at the full half of the glass without acknowledging that something is missing. For example:

• The surgeon’s experience and judgment remain among the most significant variables determining survival outcomes and complications.
• Highly subjective endoscopic evaluations are the cornerstone of entire medical fields (e.g., laryngoscopy, GI endoscopy, bronchoscopy).
• Even histopathology, often regarded as the gold standard, can suffer from a high degree of variability in its interpretation.

These are examples of substantial medical domains that will need to be improved to see practical advancements in the care of patients.
However, the first step will challenge the usual paradigm of separating scientific fields into semi-insulated compartments. The successful integration of molecular biology within the medical field is a precedent, suggesting a similar potential for a synergistic conjunction of medicine and computer science. This interdisciplinary approach is crucial for catalyzing real-life developments in patient care.

In summary, we need a new breed of doctors with the skills to understand and employ AI effectively. And this will likely require restructuring our current medical training.

A Surgeon’s Reflections on Artificial Intelligence was originally published in TDS Archive on Medium, where people are continuing the conversation by highlighting and responding to this story.

Why is the EMA’s guidance important?

The current EMA Recommendation paper on decentralised elements in EU clinical trials is a game-changer for the future of clinical research. With the COVID-19 pandemic highlighting the benefits of decentralised procedures in healthcare and clinical trials, the EMA issued guidance to manage clinical trials during the pandemic. However, the new recommendation paper goes beyond a pandemic scenario, providing guidance on the use of decentralised elements in clinical trials, regardless of any health crisis.
The overall aim of the EMA Recommendation paper on decentralised elements in EU clinical trials is to support the development of innovative solutions that not only improve patient outcomes but also ensure the necessary level of trial participant safety, protection of their rights and dignity, and the reliability of data for publication and submission for regulatory decision-making. In particular, these guidelines cover the roles and responsibilities of the sponsor and investigator, electronic informed consent, IMP delivery, trial-related procedures at home, data management, and monitoring in a DCT setting.

Having a clear and comprehensive regulatory framework for DCTs is crucial to facilitate the development and conduct of these practices. By providing explicit guidelines for executing DCTs, it will be possible to decrease the degree of uncertainty in this area, streamlining the process of developing DCT solutions and ultimately reducing the time and resources required to bring new treatments to market. Additionally, it will ensure that the rights and welfare of the trial participants are protected at all times, and that the data generated is of high quality and suitable for regulatory decision-making.

What is the role of home nursing?

One of the most significant aspects of DCTs is the use of home nursing and the delivery of trial-related procedures outside the traditional clinical trial site.
A recent article shared by the Center for Information and Study on Clinical Research Participation (CISCRP) has shown that travel is ranked the top burden for patients participating in clinical trials (31% of respondents), followed by the length of the study visits and undergoing diagnostic tests (21% of respondents). The study highlights that patients who were offered the option of study visits at their home or office were more likely to report receiving better care and attention compared to those who did not have access to such convenience-enhancing solutions. Notably, the study found that younger patients were more likely to utilise such options in clinical trials, such as visits at home or at the office. This suggests that offering more flexible and convenient options for clinical trial participation can improve patient engagement and satisfaction, particularly among younger patient populations.

The EMA guidance acknowledges the potential benefits of home nursing for clinical trials but also highlights the need to ensure the quality and safety of these services. This includes ensuring that home nursing staff have the necessary training and competencies and that appropriate systems are in place to monitor each step of the process and support patients remotely.

Addressing regulatory requirements with innovative solutions

In this context, digital solutions can help in training and oversight in home nursing for clinical trials by providing a system for training and onboarding new staff remotely, as well as ongoing professional development.
Specifically, a dedicated platform for tracking and documenting training and competencies can ensure home nursing staff are qualified and competent to perform their duties. In addition, features that allow tracking and recording of each step during trial delivery in home-based operations can improve the overall quality and safety.

With the increasing trend towards decentralisation in clinical trials, it will be important for home nursing providers to leverage these solutions to ensure that they are providing high-quality care to patients participating in clinical research, while maintaining oversight and compliance to regulations. This will not only improve the overall patient experience, but also ensure that the data generated is of the highest quality, suitable for regulatory decision-making, and ultimately help bring new treatments to market faster.

Links: https://health.ec.europa.eu/system/files/2022-12/mp_decentralised-elements_clinical-trials_rec_en.pdf
https://www.ciscrp.org/wp-content/uploads/2021/06/Sine2021_Article_PatientEngagementInitiativesIn-FINAL-Article.pdf

Find out more at www.careforme.io

Mobile research nursing