Geisinger (it is not any fiord)

A suggestion for weekend reading: excellent interview by Eric Topol with the Geisinger Health System CEO, in Medscape.

Regarding genetics in clinical practice:

Our approach is based on asking, "What if we did whole-exome sequencing on our entire population?" We started this to see whether we could find medically actionable genetic conditions that we could do something about. We are 200,000 patients into it—or I should say, research subjects, because we started with research—and we have found that 3.5%-4% of those people have something actionable: BRCA, malignant hyperthermia, Lynch syndrome, genes associated with cardiac arrhythmia.

They are unique.

And this is what happens at Sotheby's after the purchase for$1.4m

Banksy is a genius 

Dempeus sobre espatlles de gegants

Sid Mukherjee: On A.I., Longevity and Being A Digital Human

En aquest link a veritats fonamentals (ground truths) sobre longevitat, IA i els humans digitals https://open.substack.com/pub/erictopol/p/sid-mukherjee-on-ai-longevity-and?r=284kv6&utm_campaign=post&utm_medium=web trobareu una magnífica entrevista d'Eric Topol a Sid Mukherjee. Molt interessant i profund, tres quarts d'hora de podcast per a reflexionar.

Atenció al minut 35, i a tot el que diu després. En parlaré dilluns.

Al link hi trobareu també la transcripció que la podeu passar pel traductor si voleu.

Hi ha motius per al cribratge genètic poblacional?

 Ten Years of Incidental, Secondary, and Actionable Findings

Population genomic screening for three common hereditary conditions : a cost-effectiveness analysis.

El cribratge genètic poblacional s'ha estat realitzant des de fa anys en alguns entorns, molt pocs. Fa cinc anys ja vaig explicar el cas de Geisinger, que ara pertany a Kaiser Permanente. Aquí hi trobareu totes les entrades d'aquest blog sobre proves genètiques.

Els resultats en diferents entorns diuen que entre el 3 i el 5% de la població tenen alguns gens amb variants patogèniques sobre els que a data d'avui podem actuar per capgirar el seu impacte. És el que en diuen "actionable". Aquesta és la taula:

Han seqüenciat el genoma o exoma i han informat només d'un nombre determinat de gens, entre 32 i 76. I això ve d'un document de posició que explica quants gens són "actionables" i a data d'avui són 83. L'origen cal buscar-lo als Genome Wide Association Studies, quan encara hi havia confiança en que trobar un gen que provoca una malaltia permetria afrontar-la i després es va veure que això passava en molts poc casos (ara sabem que paga la pena informar només de 83, els 20.000 ).

Al NEJM ens expliquen un estudi que:

A unique aspect of this study was the reporting of life-span data for persons who carried a pathogenic or likely pathogenic variant as compared with those who did not, in an analysis involving data from 27,546 persons who had died by 2020 and life-span estimates for other persons who were at least 65 years of age. Key findings included an observed shortening in median life span among persons who carried these variants (86 years vs. 87 years for those without). Harboring such a variant in a cancer gene was associated with the largest difference in life span (3 years). By the age of 65 years, 10% of the participants with pathogenic or likely pathogenic variants in cancer genes had died. In contrast, in the group of participants without such a variant, 10% had died by 73 years of age. 

Quedem-nos amb la idea, trobar aquesta variant patogènica en un gen suposava una diferència de 10 anys de vida.  Al món hi ha dos països que fan cribratge per BRCA2, Israel (població Ashkenazy) i Islàndia. No n'hi ha enlloc més per ara. Ara bé, què costaria fer un cribratge poblacional d'això? Quin seria el cost-efectivitat?

L'Eric Topol ressenya un estudi on per cancer d'ovari i mama, Lynch sindrome i colesterolemia familiar seria cost-efectiu per:

adults younger than 40 years if the genomic test was at a relatively low cost of $250 or for people 50 years old at $166 per test (summary for patients). We are now at that level whereby whole genome sequencing has dropped to the $200 threshold, and targeted sequencing or arrays can be done at substantially lower costs. It doesn’t appear that we can continue to use cost now as the reason to not obtain information that can be lifesaving.

 Ens cal molta més informació i anàlisi d'una qüestió tant complexa. Ens trobem davant d'una situació nova que fa molt pocs anys no ens hauríem imaginat. Cal estar atents al cribratge genètic poblacional, i també a què farem quan trobem una variant patogènica "actionable", perquè si no hi ha acció (que vol dir accés a tractament cost-efectiu), llavors, per què cal trobar? Alimentaríem encara més la genòmica recreativa, que a hores d'ara interessa econòmicament moltíssim a alguns.

Human genomics vs. clinical genomics

 Today my suggestion is to read the post by Eric Topol with the same title. 

It begins with this statement:

We’re now well over 20 years since the first human genome was sequenced, but with few exceptions the massive amount of data that has been generated has not been transformed to routine patient care.

So, why?

Tackling COVID-19 beyond testing

How We Can Tackle the COVID-19 Crisis Beyond Testing

If you wear a smartwatch or fitness tracker, you can play a role in monitoring the spread of COVID-19 and other viral diseases like the flu. In this Front Row lecture, Eric Topol, MD, and Jennifer Radin, PhD, discuss how they’re calling on the public to share data from wearable devices for a study that’s helping scientists flag the early onset of contagious respiratory illnesses. By harnessing this key data—including heart rates, sleep and activity levels—from hundreds of thousands of individuals, they seek to improve real-time disease surveillance.

Medicine as a data science (3)

High-performance medicine: the convergence of human and artificial intelligence

If you want to know the current state of artificial intelligence in medicine, then Eric Topol review in Nature is the article you have to read. A highlighted statement:

There are differences between the prediction metric for a cohort and an individual prediction metric. If a model’s AUC is 0.95, which most would qualify as very accurate,
this reflects how good the model is for predicting an outcome, such as death, for the overall cohort. But most models are essentially classifiers and are not capable of precise prediction at the individual level, so there is still an important dimension of uncertainty.

And this is good summary:

Despite all the promises of AI technology, there are formidable obstacles and pitfalls. The state of AI hype has far exceeded the state of AI science, especially when it pertains to validation and readiness for implementation in patient care. A recent example is IBM Watson Health’s cancer AI algorithm (known as Watson for Oncology). Used by hundreds of hospitals around the world for recommending treatments for patients with cancer, the algorithm was based on a small number of synthetic, nonreal cases with very limited input (real data) of oncologists. Many of the actual output recommendations for treatment were shown to be erroneous, such as suggesting the use of bevacizumab in a patient with severe bleeding, which represents an explicit contraindication and ‘black box’ warning for the drug. This example also highlights the potential for major harm to patients, and thus for medical malpractice, by a flawed algorithm. Instead of a single doctor’s mistake hurting a patient, the potential for a machine algorithm inducing iatrogenic risk is vast. This is all the more reason that systematic debugging, audit, extensive simulation, and validation, along with prospective scrutiny, are required when an AI algorithm is unleashed in clinical practice. It also underscores the need to require more evidence and robust validation to exceed the recent downgrading of FDA regulatory requirements for medical algorithm approval

Therefore, take care when you look at tables like this one:

Prediction	n	AUC	Publication (Reference number)
In-hospital mortality, unplanned readmission, prolonged LOS, final discharge diagnosis	216,221	0.93* 0.75+0.85#	Rajkomar et al.96
All-cause 3–12 month mortality	221,284	0.93^	Avati et al.91
Readmission	1,068	0.78	Shameer et al.106
Sepsis	230,936	0.67	Horng et al.102
Septic shock	16,234	0.83	Henry et al.103
Severe sepsis	203,000	0.85@	Culliton et al.104
Clostridium difficile infection	256,732	0.82++	Oh et al.93
Developing diseases	704,587	range	Miotto et al.97
Diagnosis	18,590	0.96	Yang et al.90
Dementia	76,367	0.91	Cleret de Langavant et al.92
Alzheimer’s Disease ( + amyloid imaging)	273	0.91	Mathotaarachchi et al.98
Mortality after cancer chemotherapy	26,946	0.94	Elfiky et al.95
Disease onset for 133 conditions	298,000	range	Razavian et al.105
Suicide	5,543	0.84	Walsh et al.86
Delirium	18,223	0.68	Wong et al.100

1. LOS, length of stay; n, number of patients (training+ validation datasets). For AUC values:
2. *, in-hospital mortality; + , unplanned readmission; #, prolonged LOS; ^, all patients; @, structured + unstructured data; + +, for University of Michigan site.

La petjada de la COVID

 Determinants of the onset and prognosis of the post-COVID-19 condition: a 2-year prospective observational cohort study

Trobo que no s'en parla prou de la petjada que la COVID ha deixat a alguns dels nostres cossos. L'Eric Topol se n'ha preocupat força i n'ha fet aquest resum:

I on diu Spain podeu considerar que diu Catalunya, perquè es refereix a aquest article. I la troballa després de 2 anys d'estudiar la població amb Post-Covid condition PCC, és aquesta:

The study included 548 individuals, 341 with PCC, followed for a median of 23 months (IQR 16.5–23.5), and 207 subjects fully recovered. In the model with the best fit, subjects who were male and had tertiary studies were less likely to develop PCC, whereas a history of headache, or presence of tachycardia, fatigue, neurocognitive and neurosensitive complaints and dyspnea at COVID-19 diagnosis predicted the development of PCC. The cluster analysis revealed the presence of three symptom clusters with an additive number of symptoms. Only 26 subjects (7.6%) recovered from PCC during follow-up; almost all of them (n = 24) belonged to the less symptomatic cluster A, dominated mainly by fatigue. Recovery from PCC was more likely in subjects who were male, required ICU admission, or had cardiovascular comorbidities, hyporexia and/or smell/taste alterations during acute COVID-19. Subjects presenting with muscle pain, impaired attention, dyspnea, or tachycardia, conversely, were less likely to recover from PCC.

Table 4Characteristics of the post-COVID-19 condition clusters.

	Cluster A	Cluster B	Cluster C
N (%)	139 (40.8)	152 (44.6)	50 (14.2)
Age, years, median (IQR)	50 (42–57)	47 (38–56)	45 (39–51)
Sex, female, N (%)	73 (52.5)	120 (78.9)	45 (90)
Hospitalization, N (%)	65 (46.8)	52 (34.2)	13 (26)
Intensive care, N (%)	10 (7.2)	5 (3.3)	1 (2)
Comorbidities, N (%)
Allergy	31 (22.3)	56 (36.8)	20 (40)
Obesity	32 (23.0)	39 (25.7)	13 (26.0)
Dyslipidemia	30 (21.6)	43 (28.3)	9 (18.0)
Hypertension	34 (24.5)	25 (16.4)	8 (16.0)
Lung disease	21 (15.1)	31 (20.4)	6 (12.0)
Persistent symptoms, N (%)
Fatigue	100 (71.9)	139 (91.4)	48 (96.0)
Neurocognitive complaints	62 (44.6)	128 (84.2)	46 (92.0)
Dyspnea	45 (32.4)	131 (86.2)	44 (88.0)
Headache	41 (29.5)	103 (67.8)	44 (88)
Myalgia	31 (22.3)	98 (64.5)	33 (66)
Arthralgia	39 (28.1)	92 (60.5)	47 (94)
Chest pain	31 (22.3)	76 (50.0)	45 (90)
Tachycardia	19 (13.7)	83 (54.6)	39 (78.0)
Cough	18 (12.9)	64 (42.1)	19 (38.0)
Neurosensitive symptoms	31 (22.3)	66 (43.4)	40 (80.0)
Diarrhea	14 (10.1)	59 (38.8)	26 (52.0)
Low grade fever	13 (9.35)	36 (23.7)	21 (42.0)
Smell alterations	34 (24.5)	38 (25)	30 (60)
Dermatological alterations	24 (17.3)	36 (23.7)	43 (86)
Dysphagia	9 (6.47)	18 (11.8)	27 (54)
Dysphonia	10 (7.19)	29 (19.1)	19 (38)
Recovery from PCC, N (%)	24 (17.3)	1 (0.7)	1 (2.0)

I la petjada després de dos anys persisteix encara en bona part de la població amb PCC. I moltes coses que encara no sabem... 

Books of the year

The Economist
Financial Times
Prospect (economics)

Top ten books of 2019 by Eric Topol

Revisant l'assistència sanitària basada en el valor

Building on value-based health care: Towards a health system perspective

En una entrevista que li van fer a l'Eric Topol, li preguntaven que en pensava de l'assistència sanitària basada en valor, i va dir que era una broma, posem-ho en context:

It’s a joke, value-based care. Basically, we have one-third of the healthcare, but $3.6 trillion is waste—low-value care. We need to stop that. That’s part of why it’s so costly. And so this whole idea of value-based care doesn’t even get to it. There’s a long list of hundreds of things that each of the professional societies have called out as being shouldn’t be done anymore. And we’re doing it every day, you know, thousands, hundreds of thousands of times, every day and week in this country. We have to get rid of the waste and inappropriate and unnecessary care and we haven’t done anything to do that here of note.

Certament, l'ús del terme value-based se n'ha anat de les mans. En Peter Smith et al. tracten de posar ordre al terme i diuen:

We therefore define health system value to be the contribution of the health system to societal wellbeing. The health system is expected to contribute to wellbeing in a number of respects, which are often expressed as a set of objectives for the health system. There is a core cluster of objectives, developed from the World Health Report 2000 , that has secured widespread acceptance amongst policy-makers as reflecting their central priorities: health improvement, responsiveness, financial protection, efficiency and equity. It is these strategic goals that should reflect the health system’s concept of value.

 i aquestes són les dimensions

Em sembla bé, però no afegeix massa al que ja coneixem de l'OCDE relatiu a Health Systems Performance Assesment.

Book choice

Everybody Wants to Go to Heaven but Nobody Wants to Die: Bioethics and the Transformation of Health Care in America

Eric Topol comment in Nature

IA pertot arreu (2)

 Foundation models for generalist medical artificial intelligence

La revista Nature publica aquesta setmana un article de revisió que explica l'estat de situació de la Intel·ligència Artificial mèdica generalitzada GMAI. Fa unes setmanes explicava en una entrada els models multimodals, i ara anem més enllà. El concepte generalista ens condueix cap a models que van més enllà de tasques, i que aprenen del context, modelitzen el llenguatge i introdueixen l'aprenentatge contrastat. Per fer-nos una idea, fins ara tot els prop de 500 models d'IA que ha aprovat la FDA eren orientats a 1 o 2 tasques. El salt cap a la IA generalista suposa un canvi substancial. I això ha passat amb pocs mesos de diferència.

Les tres capacitats clau:

First, adapting a GMAI model to a new task will be as easy as describing the task in plain English (or another language). Models will be able to solve previously unseen problems simply by having new tasks explained to them (dynamic task specification), without needing to be retrained3,5. Second, GMAI models can accept inputs and produce outputs using varying combinations of data modalities (for example, can take in images, text, laboratory results or any combination thereof). This flexible interactivity contrasts with the constraints of more rigid multimodal models, which always use predefined sets of modalities as input and output (for example, must always take in images, text and laboratory results together). Third, GMAI models will formally represent medical knowledge, allowing them to reason through previously unseen tasks and use medically accurate language to explain their outputs.

a, A GMAI model is trained on multiple medical data modalities, through techniques such as self-supervised learning. To enable flexible interactions, data modalities such as images or data from EHRs can be paired with language, either in the form of text or speech data. Next, the GMAI model needs to access various sources of medical knowledge to carry out medical reasoning tasks, unlocking a wealth of capabilities that can be used in downstream applications. The resulting GMAI model then carries out tasks that the user can specify in real time. For this, the GMAI model can retrieve contextual information from sources such as knowledge graphs or databases, leveraging formal medical knowledge to reason about previously unseen tasks. b, The GMAI model builds the foundation for numerous applications across clinical disciplines, each requiring careful validation and regulatory assessment.

Tres aplicacions potencials:

 a, GMAI could enable versatile and self-explanatory bedside decision support. b, Grounded radiology reports are equipped with clickable links for visualizing each finding. c, GMAI has the potential to classify phenomena that were never encountered before during model development. In augmented procedures, a rare outlier finding is explained with step-by-step reasoning by leveraging medical domain knowledge and topographic context. The presented example is inspired by a case report58. Image of the fistula in panel c adapted from ref. 58, 

L'aplicació de la IA a la medicina representa una impuls transformador crucial, i encara no en sabem les conseqüències. L'article detalla alguns dels reptes, cal llegir-lo per fer-nos una idea de cap on va la cosa. 

 PS. Per entendre el context de l'article, Eric Topol al seu blog.

 

Ownership and access to medical data

Unpatients—why patients should own their medical data

Eric Topol says in Nature Biotechnology:

Today, in the United States, health data live in a plethora of places, from electronic health record (EHR) systems, insurance claims databases, siloed personal health apps, research and clinical trial databases, imaging files and lots of paper. Although seemingly everywhere, any true semblance of an overarching organization or standardization of medical data are lacking, whether at the individual or societal level

His proposal is straightfoward: the ownership of the clinical record is of the patient. This situation is completely different in our country. We have public centralised repositories and the patient is the owner. There is still a lack of coordination and many things to solve, however the basics are covered in the publicly funded System, that's not the case in the private sector.

In contrast to the legal and technical difficulty an individual faces to obtain all his or her own medical data is the relative ease with which hackers have managed to breach ~100 million patient records in the first half of 2015

And his proposal:

 We must begin talking about creating a health data resource in a much broader and more universal context, controlled by the individuals who supply the data. This is a unique moment where we may be able to provide for personal control and, at the same time, create a global knowledge medical resource.

Sounds interesting, though methodology is crucial for success.


PS. Hacking electronic records:

The timeline for electronic medical data hacks in the United States of over 1 million individuals

Transforming the practice of care in the most inefficient and wasteful health system

The Smart-Medicine Solution to the Health-Care Crisis

Eric Topol provides clear insights for a wide range of life sciences issues, and some days ago he insisted once again on the need to reform US health system. Everybody is talking about financing and acces, and he focuses on organization. That's good to hear. I suggest a close look at the WSJ article. Although the scope is US, you'll find many comments that are absolutely useful for our health system (the public and specially the private one).

Our health-care system is uniquely inefficient and wasteful. The more than $3 trillion that we spend each year yields relatively poor health outcomes, compared with other developed countries that spend far less. Providing better health insurance and access can help with these problems, but real progress in containing costs and improving care will require transforming the practice of medicine itself—how we diagnose and treat patients and how patients interact with medical professionals.

And he backs a smart medicine practice:

Smart medicine offers a way out, enabling doctors to develop a precise, high-definition understanding of each person in their care. The key tools are cheaper sensors, simpler and more routine imaging, and regular use of now widely available genetic analysis. As for using all this new data, here too a revolution is under way. 

And the key integrative tool:

At the Scripps Research Institute, we are working with the support of a National Institutes of Health grant and several local partners to develop a comprehensive “health record of the future” for individual patients. It will combine all the usual medical data—from office visits, labs, scans—with data generated by personal sensors, including sleep, physical activity, weight, environment, blood pressure and other relevant medical metrics. All of it will be constantly and seamlessly updated and owned by the individual patient.

Good news (US only):

 Fortunately, serious ventures in smart medicine are well along. My colleagues and I at the Scripps Research Institute are leading the Participant Center of the NIH’s Precision Medicine Initiative, which is currently enrolling one million Americans. Volunteers in the program will be testing many of the new tools I have described here. The recently formed nonprofit Health Transformation Alliance, which includes more than 40 large companies providing health benefits to 6.5 million employees and family members, intends to address the high cost of health care by focusing on, among other things, the sophisticated use of personal data.

I have to say that his position is well grounded, it is not a fascination for technology. The true health reform starts with the practice of medicine. Completely agree.

Fotopres

Covid-19 testing landscape

COVID-19 diagnostics in context

This is the best summary of current supply of diagnostic tests for Covid-19:

COVID-19 tests can be grouped as nucleic acid, serological, antigen, and ancillary tests, all of which play distinct roles in hospital, point-of-care, or large-scale population testing.
Table 1 summarizes the existing and emerging tests, current at the time of writing (May 2020). A continuously updated version of this table is available at https://csb.mgh.harvard.edu/covid

Eric Topol says:

There are now *88* @US_FDA  cleared (by EUA) #COVID19 tests so far. Their false negative rates range from 10-48% (by post-release reports).
Might be better to have less tests, more accuracy, with faster turnaround

I agree.

Table 1 Performance comparison of different test types.
Throughput is determined by process type and assay time. In general, automated plate-based assays have higher daily throughputs. Hashtag (#) indicates example systems that have received FDA emergency use authorization (FDA-EUA). See https://csb.mgh.harvard.edu/covid to access continuously updated information. PCR, polymerase chain reaction; PCR-POC, PCR–point-of-care; ddPCR, digital droplet PCR; NEAR, nicking endonuclease amplification reaction; RCA, rolling circle amplification; SHERLOCK, specific high-sensitivity enzymatic reporter; DETECTR, DNA endonuclease-targeted CRISPR transreporter; NGS, next-generation sequencing; μNMR, micro–nuclear magnetic resonance; LFA, lateral flow assay; ELISA, enzyme-linked immunosorbent assay; CLIA, chemiluminescence immunoassay; EIA, enzyme immunoassay; ECLIA, electrochemiluminescence immunoassay; ECS, electrochemical sensing; VAT, viral antigen assay; IFM, immunofluorescence microscopy; WB, Western blot.

Type	Target	Virus	Assay time	Process type	FDA-EUA	Examples
PCR	Viral RNA	SARS-CoV-2	2–8 hours; >12 hours	Plate	56	#Roche, #LabCorp, #BioMerieux, #Qiagen, #Perkin-Elmer, #Becton Dickinson, #Luminex, #Thermo Fisher, others
PCR-POC	Viral RNA	SARS-CoV-2	<1 hour="" td="">	Cartridge	2	#Cepheid, #Mesa, Credo
ddPCR	Viral RNA	SARS-CoV-2	2–4 hours	Manual	1	#BioRAD
NEAR	Viral RNA	SARS-CoV-2	15 min	Cartridge	1	#Abbott
OMEGA	Viral RNA	SARS-CoV-2	1 hour	Plate	1	#Atila BioSystems
RCA	Viral RNA	SARS-CoV	2 hours		0
SHERLOCK	Viral RNA	SARS-CoV-2	1.5 hours	Kit	1	#Sherlock Biosciences (CAS13a)
DETECTR	Viral RNA	SARS-CoV-2	1 hour	Kit	0	Mammoth Biosciences (CAS12a)
NGS	Viral RNA	SARS-CoV-2	Days		1	#IDbyDNA, Vision, Illumina
μNMR	Viral RNA	SARS-CoV-2	2 hours	Cartridge	0	T2 Biosystems
LFA	IgG, IgM	SARS-CoV-2	15 min	Cartridge	3	#Cellex, #Sugentech, #ChemBio, Innovita
ELISA	IgG, IgM	SARS-CoV-2	2–4 hours	Plate	4	#Mount Sinai, #Ortho-Clinical (2), #EUROIMMUN US Inc., BioRAD, Snibe, Zhejiang orient, Creative Dx
CLIA	IgG, IgM	SARS-CoV-2	30 min	Cartridge	2	#Abbott, #DiaSorin
EIA	IgG, IgM	SARS-CoV-2	2 hours	Plate	1	#BioRAD
MIA	IgG, IgM	SARS-CoV-2		Plate	1	#Wadsworth Center
ECLIA	IgG, IgM	SARS-CoV-2	20 min	Plate	1	#Roche
ECS	IgG, cytokine	SARS-CoV-2	1 hour	Cartridge	0	Accure Health
VAT	Viral antigen	SARS-CoV-2	20 min	Cartridge	1	#Quidel, Sona NT, RayBiotech, SD Biosensors, Bioeasy
Microarrays	Ig epitopes	SARS-CoV-2	1.5 hours	Plate	0	RayBiotech, PEPperPRINT
IFM	Viral protein	SARS-CoV	3 hours	Manual	0
WB	IgG, IgM; viral protein	SARS-CoV	4 hours	Manual	0