Brendan Dunphy
Passionate about the opportunities to apply AI in healthcare to improve patient outcomes and effciency, in particular for rare diseases.
Location Based in Nice, France and with business and personal interests in the US and UK.
Activity
-
Many thanks @RomainNicolle, I will take a look. And a BIG thankyou for all your prompt and insightful answers and comments over the last few weeks - chapeau!
-
Brendan Dunphy made a comment
Probably one of the best online courses I have taken and this is mainly down to the quality facilitation, a common, weakness of online education - chapeau!
If you can find a way to incorporate more graphics to lighten the textual burden, actively link to definition of terms and lighten week 3 then it would be perfect.
-
@RoselineFavresse TFS. 'The average cost of rare diseases was again approximately 10x higher than for mass market diseases.' Besides the human and social arguments there also appears to be an economic one!
-
Brendan Dunphy made a comment
The implications of a diagnosis on insurance is not something I have thought about. Can you provide any references?
-
Brendan Dunphy made a comment
"first, do no harm" might be more understandable by most than than 'non-maleficence'?
-
I'm curious to know if anyone has direct experience of a patient recording (audio or video) a consultation with a geneticist (or indeed any clinician!) to play back later. It is legal to do so in the UK, but rare, and not encouraged by the medical profession.
-
These are great initiatives.
-
Yes, and to find better alternatives where they exist.
-
Brendan Dunphy made a comment
I came across pharmacogenomics a few months ago and urge anyone suffering with severe side effects to treatments to explore whether there is help available. Also, an interetsing research article 'Exome sequencing allows detection of relevant pharmacogenetic variants in epileptic patients' at https://www.nature.com/articles/s41397-022-00280-w (paywall)
-
Brendan Dunphy made a comment
Any other examples where these tools are being deployed in practice elsewhere?
-
Brendan Dunphy made a comment
A useful summary of multiomics in the link
-
Brendan Dunphy made a comment
The need to better integrate genetics into rare disease diagnostics stand out.
-
You are not alone @AnaMariaRamirez-Scarlett !
-
Brendan Dunphy made a comment
I can see graph databases playing a huge role in capturing genomics data and making it available for both human visualisation and interpretation and machine computation.
-
Brendan Dunphy made a comment
The traditional separation between the clinical and genetic domains in that journey is understandable but clearly a problem going forward. By bringing genetic testing and knowledge earlier into the journey it should be possible to reduce the iterations, costs and time to diagnosis for more patients.
-
FYI, the Australia link is returning a 404 error
-
Brendan Dunphy made a comment
Clarification please. So 50% of patients who go through full exome testing remain with no diagnosis?
-
Brendan Dunphy made a comment
I can see that collecting more precise and consistent phenotyping data is a challenge for physicians and patients. It takes time, tools, training, validation etc. I expect advances in genetics may well be constrained by this.
-
Brendan Dunphy made a comment
If not a futurelearn platform limitation, it would be useful for all terms to be linked to their definition, ideally visible on hover.
-
Brendan Dunphy made a comment
FYI, the word 'software' is an uncountable noun as it’s a collective noun or a mass noun (another exaple is 'sheep'!). So the plural form of software would still be spelled the same way as it is in its singular form “software”. This is true if you are talking about different types of software or multiple programs open at the same time.
-
Brendan Dunphy made a comment
It becomes ever more clear that we need to do better at routinely and consistently capturig the data of every undiagnosed patient if we are to discover all pathogenic variants.
-
A fascinating thread. In medicine 'guidelines' created and updated by panels of experts from a specilaist domain or professional body are common. They exist to capture and share best practice, improve outcomes, reduce variance etc. Is there an equivalent in genetics or is it too soon? And if they exist, can you point me to an example? (Forgive me if this has...
-
Review is indeed critical and I know of some hospitals that are formalising this process across all their undiagnosed patient records and using AI to surface patients that may benefit from new research knowledge not previously available at their last encounter.
-
Brendan Dunphy replied to Ana Pataki
FYI, the NHS implemented a federated platform for Covid data sharing and that is now being extended for wider use cases and currently under procurement, though delayed - https://www.digitalhealth.net/2022/04/nhs-england-federated-data-platform/ To my knowledge, genetic data is not (yet) a projected primary use case
-
Brendan Dunphy made a comment
A challenging week but thanks to the faculty for prompt and insightful responses. This is not usually the case with online courses so kudos!
-
Could you expand on what you mean by the last sentence @EvanGOUY - what are the exact issues - legal, ethical, practical?
-
We need to see parallel improvements in clinical data gathering and availability at point of need if we are to meet this challenge. Although digital patient records are increasingly common the data they contain is far from complete, upto date or accurate, and often largely unstructured and in note form. I hope that the increasing availability of NGS can be...
-
I found the Next Gen pdf useful
-
They barely exist in some parts of the world @AnaMariaRamirez-Scarlett, and even in rural and other areas in France and elsewhere. And where they do exist they are often over-stretched as a recent report from the GMC noted with some GP's (primary care) now seeing 3x the agreed safe limit per day! If we and they are to benefit from NGS then all health systems...
-
Brendan Dunphy replied to Ana Pataki
Probably also worth mentioning synthetic data as a way to compensate for the lack of available data and privacy concerns for training algorithms. See https://blogs.nvidia.com/blog/2021/06/08/what-is-synthetic-data/. It's still early days but Pharma are active users today. In essence it's a way to scale-up or flesh out limited data sets and ensure the resulting...
-
Brendan Dunphy replied to Romain Nicolle
I have no doubt that AI will do most of the 'heavy lifting' in knowledge intensive araes such as medicine - it will be faster, more accurate, more consistent, work 24/7 and way cheaper! However, the most important part of the process is communication with the patient. I don't see that need going away (but rather increase) and hopefully AI can create more time...
-
Brendan Dunphy replied to Brendan Dunphy
@RomainNicolle Thanks!
-
Well I have to agree with that @RomainNicolle In my mind there is only one true universal language, and that is mathematics!
-
Brendan Dunphy replied to Pascal C
@RomainNicolle Most welcome!
-
Brendan Dunphy replied to Brendan Dunphy
Yes, indeed it is @RomainNicolle. In your opinion, is sufficient effort and coordination in-place or is there a need to acclerate and fund this coopertaion?
-
It certainly is @RomainNicolle !
-
Brendan Dunphy made a comment
Are these videos in the public domain and free to share, with credits of course?
-
Brendan Dunphy made a comment
I'm pleased to hear Dr Byunner say this given what I perceive as a 'lingering scepticism' as to the role of AI in diagnostics by some.
'I think AI is superior to human knowledge, memory, creativity in the arena of diagnostics. So, I have very little doubt that in 5 years AI will be the main stay of all diagnostics.'
-
Brendan Dunphy made a comment
I assume this is not the case anywhere today so a call for action? 'In summary, academic and commercial laboratories must share their data, contribute and track the variant classification to be able to transfer the real-time updated classification of a variant to the patients and their physicians.'
-
Brendan Dunphy made a comment
What is meant by 'human population groups' in the above text?
-
Brendan Dunphy made a comment
Setting patient expectations is key and we need to do more to lower expectations given the immaturity of the science and the media hype and misinformation around genetics. This includes how long it takes to get results. I know of 'low priority' cases in the UK where results may take a year rather then the 'few weeks' mentioned on NHS web sites. This is a long...
-
Brendan Dunphy replied to Pascal C
FYI and from the UK's NHS 'The offer to take part in the NGRL (National Genomic Research Library) should be made to all patients and family members being offered WGS. However, this discussion may take place at a later point after clinical WGS has been initiated, if clinically appropriate.' If approval is given for NGRL research purposes then the retained...
-
Maybe consider the use of graphics to make this information easier to visualise and digest? Solething along the lines of this from the UK's NHS genomics education - https://www.genomicseducation.hee.nhs.uk/wp-content/uploads/2021/06/Genomic-Testing-Infographic_w-title.pdf
-
Tx @RomainNicolle In the US it seems to be of most value to primary care and other physicians rather than genetists. I think there is a big need for AI/ML RD identification tools and related education in primary care given it is often the gateway to diagnosis.
-
Brendan Dunphy made a comment
Could someone expand on the reasoninbg for this sentence above? 'A huge international effort should be put in databases reporting the pathogenicity of variants in non-coding DNA.'
-
@MagdaGranata I look forward to that!
-
Sure, AI and ML will never replace all aspects of the daignostic process but they can 'augment' the clinicains during that process, including the physical examination. https://www.face2gene.com/ is a good example of this today
-
-
Brendan Dunphy made a comment
Reconsent can be a major logistical challenge. I know Rare X is facing this for some third-party patient cohorts.
-
@CaroleChehowah Given the trajectory of computing and AI I suspect that this will one day be the reality, though not in my time!
-
Brendan Dunphy made a comment
What type of deafness? What type of deafness? There are a lot of variations to confound a diagnosis.
-
The HPO link is not working for me but this one does - https://hpo.jax.org/app/about
-
I think it's also worth noting that even if a legal right to access and share medical records exists (both within and across country's) there are many practocal challenges to do so as standards and formats vary making it challenging to aggregate and 'normalise' the records. In the US this right exists for several years but in practice it is still...
-
Brendan Dunphy made a comment
Beautifully expressed in all its pain and love.
-
I really think we can do better in creating a 'longitudinal' patient record that captures, summarises and acts as an index to, the patient's journey. Fragmentation of data and encounters is truly a root cause of delays, errors and omissions in diagnosis. We need to make it easier for clinicians to see the big picture and the missing pieces, reduce repetition...
-
@PascalC I think there is often confusion between medical and 'recreational' testing in France with the former largely legal but the latter (e;g. for geaneology) still illegal, I think the only country in Europe where this is the case?
-
Newborn screening is now common in Europe too with The Netherlands having one of the most comprehensive programs.
-
@RoselineFavresse I note that the survey above is now closed - would you happen to have a link to the results?
-
@CristinaSkrypnyk you make a strong case as to why patients in general (and RD patients in particular) should 'own' all their data and be able to easilly share - for diagnosis, care and research - as they see fit.
-
Trends in technology (mobile, distributed, cloud etc) make personal ownership a possibility and there are clear trends in this direction. Hospital IT systems are notoriously poor for clinical use and primarilly designed for patient and facility adminsitration, billing etc. Exchaning patient data between systems can also be a challenge but varies across...
-
For many, not knowing is not a benign state. They either struggle for medical attention or are subjected to endless tests and mis-diagnosis over many years. I can only imagine the relief that comes with knowing and the end of that odyssey.
-
Lol....and this is a BIG barrier to the systemic use of technology to help organise, store and access medical data in general, and rare disease data in particular. The good news is that we are making progress but we do need to keep working on improving and standardising the language and its use to improve our ability to communicate, share and compute.
-
Brendan Dunphy replied to Laura Valinotto
This is a great testament to what can be done - well done!
-
along th education, I think improving access to the evidenced-based information that already exists at the 'point-of-need' is critical to this challenge, and not only in Mexico, but every country. This information and data may be incomplete and imperfect in many ways but it should be available digitally and in ways that are easy to consume by both primary care...
-
Brendan Dunphy made a comment
Hello everyone. Great to see such an international cohort!
I'm the co-founder of a US-based company focused on using Artificial Intelliegence (AI) and Machine Learning (Ml) to improve clinical diagnosis, including for rare diseases, where we already have a functioning prototype called ENOLA.
My colleagues are clinicians (so more than a few steps ahead...
-
Brendan Dunphy made a comment
I think the potential is positive but I doubt that policy makers or citizens have fully grasped the implications and there remain many barriers to its use beyond selling us more useless stuff!!!
-
Brendan Dunphy made a comment
I would like to see the use of Big Data for our collective good rather than individual or commercial gain, the predominant focus today.
-
Brendan Dunphy made a comment
I suspect only a very small % of users have ever read the T&C's related to accepting cookies! GDPR is well intended but overly cumbersome for consumers. There also needs to be a simple way to both 'see' ,and control use of ones own data.
-
Brendan Dunphy made a comment
It's application to non EU citizens in the EU
-
Brendan Dunphy made a comment
Forecasting electricity usage across national grid networks based on weather forecasts, holidays, prior usage patterns, global events etc.
-
Brendan Dunphy made a comment
Remove outliers?
-
Brendan Dunphy made a comment
NHS trust fast-tracks AI tool which allows doctors to view scans remotely. See https://www.digitalhealth.net/2020/05/nhs-trust-ai-tool/ I note:
"Dr Kiruba Nagaratnam, consultant stroke physician and geriatrician, and clinical lead for stroke medicine at the trust, said: “This work has revolutionised the way we traditionally reviewed scans and made...
-
Brendan Dunphy made a comment
Depends on the uses and who would have access, and maybe for how long.
-
Brendan Dunphy made a comment
ACD - excel in one of 3 domains
-
Brendan Dunphy made a comment
Sometimes 'more is less' and I think professional sports (already over inflated) is one case where this is true. I know 'optimisation' is one inevitable application domain for Big Data but I prefer others that generate new insights and help achieve public good
-
Brendan Dunphy made a comment
Monitor the spread of pests and diseases such as tiger mosquitoes
-
Brendan Dunphy made a comment
data access (from sensors) and heterogeneity of data (different research domains and experts) stand out for me
-
Brendan Dunphy made a comment
Need quasi-real time data feeds from satelites
-
Brendan Dunphy made a comment
Big data in health
-
Brendan Dunphy made a comment
Big data in Health domains
-
Brendan Dunphy made a comment
I like the structure of the course. I wonder how valuable it will be. I would like to know more about risks, failures and essentials for success.
-
Brendan Dunphy made a comment
LinkedIn use most of these esp. demographic, location, intention and behavioral, for recruitment and career developement dervices
-
Brendan Dunphy made a comment
Strava could limit access to certain data or provide this option to users.
-
Brendan Dunphy made a comment
The table is 'lean' data, focused and to-the-point, and the infographie more à story.
-
Brendan Dunphy made a comment
I think the first step is to identify the need or question to be answered and this should be explicit.
-
Brendan Dunphy made a comment
My data is fragmented across hundreds of places. Some is visible to me but most is not. I have no idea of my big data 'asset', how to view, control or value it.
-
Brendan Dunphy made a comment
I've been contributing from the beginning, both work and pleasure, but not so much social media, too invasive and addictive for me.
-
Brendan Dunphy made a comment
All the explicit and implicit data I create as I navigate my digital world
-
Brendan Dunphy made a comment
I have decreasing faith in Government and Business desire and ability to protect our data and respect our privacy
-
Brendan Dunphy made a comment
Hello. Brendan from France. Here to validate my knowledge and decide if to dig deeper into the topic.
-
Brendan Dunphy made a comment
Interesting to see a UK algorithm being adopted in the USA before the UK - https://www.aiin.healthcare/topics/medical-imaging/ai-model-pneumothorax-fda-clearance?utm_source=newsletter&utm_medium=ai_news I expect this will be the pattern given differing risk appetites and healthcare structures
-
Many thanks @AllanPang, and to your wife.
-
Tx @NanaM Me too, getting them 'to stick' is the challenge! I find it easier when visualised as a model rather than text.
-
@NanaM Tx - a topical example! So Recall is its overall accuracy and Precision its accuracy in identifying positive cases?
-
Tx @PaulBromiley I still think this is too theoretical for most and we need to think about what diagnostic and prescribing artefacts that encapsulate key transitional data between a patient and clinician can be routinely shared, and how. Maybe mobile apps are a useful part of this?
-
Agreed; I think the interesting point was the relative value of two sets of eyes over one!
-
Brendan Dunphy made a comment
Thanks!
-
Totally agree, especially the second point. We need to spend a lot more time on building user interfaces that as intuitive as the best out there (think Apple, mobile social media etc) and increasingly, voice (hands free). Today we have interfaces designed for offices poorly adapted for use in healthcare.
-
This illustrates the PULL where patients are requested to use a device to provide data but what about the PUSH where patients are using devices and collecting data on their own initiative - should this also be integrated into the existing system and if so, how?
-
Interested to know how many UK GP surgeries do not have a sysem such as System 1 mentioned in the vid - any?
