Your SlideShare is downloading. ×
2025: 20% doctor included? an exercise in technology speculation & musings
Upcoming SlideShare
Loading in...5

Thanks for flagging this SlideShare!

Oops! An error has occurred.

Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

2025: 20% doctor included? an exercise in technology speculation & musings


Published on

Published in: Health & Medicine
  • Be the first to comment

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

No notes for slide
  • Causes 40–80,000 deaths annuallyArticle: Graber - Bringing Diagnosis Into the Quality and Safety Equations ( errors in diagnosis stem from mistakes in the interpretation of diagnostic tests. For example, pathology, radiology, and the clinical laboratory each have error rates of 2% to 5%. Superimposed on these testing errors are the ubiquitous system-related errors encountered in every health care organization, as well as cognitive errors caused by faulty clinical reasoning. Diagnostic errors do not occur only in connection with unusual conditions but span the breadth of clinical medicine, from rare disorders to commonplace ones like anemia and asthma.
  • Rivals the number of deaths from breast cancerArticle: Winters – Diagnostic errors in the intensive care unit: a systemic review of autopsy studies ( investigators, led by Bradford Winters, MD, PhD, examined studies that used autopsy to detect diagnostic errors in adult ICU patients. In 8% of patients, the diagnostic error was serious enough that it may have caused or directly contributed to the patient's death, and had doctors been aware of the proper diagnosis, the treatment likely would have been different.
  • 90 cases involved injury, including 33 deathsArticle:
  • Article:
  • Premature closure - The failure to continue considering reasonable alternatives after an initial diagnosis was reached.Article:
  • Source:|74|305|339|342Dr. Jonathan Schaffer - Managing director of the Cleveland Clinic's MyConsult Online Medical Second Opinion serviceMyConsult also began offering its services to large, self-insured employers when it launched in 2002, said Dr. Schaffer, “but in the last couple of years, the smaller employers are becoming more interested, so we now have employers with as few as 500 employees.”While the majority of the employers using MyConsult are self-insured, some insured employers also are making the service available to employees case by case. The cost averages about $565 per case, with an additional $180 charged if pathology is needed, he said.Dr. Schaffer said Cleveland Clinic doctors disagree with initial diagnoses in 11% of cases. In 15% of cases, they find the need for further testing. In 22% of the cases reviewed, minor changes to treatment plans are recommended, moderate changes were recommended in 26% of cases and major changes were recommended in 18% of cases.
  • Source: number of recommendations and the distribution of classes of recommendation (I, II, and III) and levels of evidence (A, B, and C) were determined. Considering the 16 current guidelines reporting levels of evidence, only 314 recommendations of 2711 total are classified as level of evidence A (median, 11%), whereas 1246 (median, 48%) are level of evidence C. Level of evidence significantly varies across categories of guidelines (disease, intervention, or diagnostic) and across individual guidelines. Recommendations with level of evidence A are mostly concentrated in class I, but only 245 of 1305 class I recommendations have level of evidence A (median, 19%).Recommendations issued in current ACC/AHA clinical practice guidelines are largely developed from lower levels of evidence or expert opinion. The proportion of recommendations for which there is no conclusive evidence is also growing. These findings highlight the need to improve the process of writing guidelines and to expand the evidence base from which clinical practice guidelines are derived.
  • Source: Surgeons given written descriptions of surgical problems have split down the middle regarding whether to recommend surgery – half recommending surgery, half not. When surveyed again 2 years later, the same surgeons often disagreed with their previous opinions, with as many as 40% changing their recommendations.
  • FYI – Requested to not disclose study detailsContext: Experts commonly warn about the risks of xxx. A panel of experts were polled for their estimates of the chances of a spontaneous yyy of an xxxx. There was large variability among the various experts’ opinions, showing that, even for something as easily measureable/trackable as xxx, practitioners of modern medicine widely disagree.
  • FYI – Requested to not disclose study detailsContext: Survey was done to get docs’ opinions on how useful screening for colon cancer is in terms of early diagnosis and prevention of death from colon cancer. Answers were roughly uniformly distributed from 0% to 100%, indicating the lack of any meaningful agreement/consensus around efficacy of a standard procedure commonly recommended in modern medicine.
  • Source: the background on fever, I pulled this from the discussion section of the article which New Scientist references which gives a pretty comprehensive review of the previous literature on antipyretics: Despite limited data suggesting a beneficial role of fever in the host response, antipyretic therapy is employed commonly to treat febrile critically ill patients. The common justificationsfor such therapy include improved patient comfort, reduction in cardiovascular stress, and avoidance of increased oxygen consumption. Although these considerations may seemreasonable, there is a surprising lack of data to support the use of antipyretic therapy for these indications. The only clinical scenario where aggressive antipyretic therapy is supported inthe literature is acute brain injury, where elevated brain temperatures appear to worsen neurological outcome. Temperature elevation has been shown to have a number of positive effects on immune function in vitro. These include increased antibody and cytokine production, enhanced neutrophil and macrophage function, increased Tlymphocyte proliferation, enhanced anti-viral and anti-tumor activity of interferons, and direct inhibition of microbial growth. Furthermore,a number of studies have suggested a beneficial effect of fever in clinical infection. These include studies from the pre-antibiotic era which demonstrated enhanced resistance to infection in febrile patients. Two previous prospective, observational studies of fever in the ICU suggested that fever is deleterious for intensive care unit patients. Circiumaru et al. studied 100 consecutive patients and found that fever alone was not associated with higher mortality, but that prolonged fever greater than five days was associated with higher mortality. Barie et al. studied 634 consecutive patients and found that peak temperature, but not infectious fever, was independently associated with mortality. Both of these studies, however, are limited by their observational design and did not compare the effects of antipyretic treatment to control patients. The only previous prospective, randomized study of antipyretic therapy in critically ill patients demonstrated no significant differences in recurrence of fever, incidence of infection,antibiotic therapy, intensive care unit and hospital length of stay, or mortality. This study was limited by small sample size and more importantly by the fact that the only antipyretictherapy employed was external cooling, certainly not the standard of care today.
  • IOM – Health Literacy: A Prescription to End Confusion (
  • Source: A failure of the assumption has immense implications for the quality of care. It implies that the same patient can go to different physicians, be told different things, and receive different care. No doubt some of the differences will not be important. However, some will surely be important—leading to different chances of benefits, different harms, and different costs. A failure of the assumption also has immense implications for informed consent, expert testimony, consensus development, the concepts of "standard and accepted" or "reasonable and necessary," malpractice, quality assurance programs that are based on statistical norms, and the cost of care.
  • Article: John P. A. Ioannidis - Why Most Published Research Findings Are False (PLoS Med, 2005) ( is increasing concern that most current published research findings are false………a research finding is less likely to be true when the studies conducted in a field are smaller; when effect sizes are smaller; when there is a greater number and lesser preselection of tested relationships; where there is greater flexibility in designs, definitions, outcomes, and analytical modes; when there is greater financial and other interest and prejudice; and when more teams are involved in a scientific field in chase of statistical significance. Simulations show that for most study designs and settings, it is more likely for a research claim to be false than true. Moreover, for many current scientific fields, claimed research findings may often be simply accurate measures of the prevailing bias
  • CHAMP IDiagnostic accuracy of the DARES core knowledge engine in a selection of acute careproblems under extremely unfavorable conditions.Imposed limitationslimited to scope of practice of medical assistants (MAʼs).limited only to past medical history, history of present illness, routine vital signs.prohibited from using physical exam findings, labs or imaging.Results (review of blinded physician assessments and final diagnosis review)91% had correct primary diagnosis as top choice.8% had correct primary diagnosis included in the differential diagnosis (required additional data for confirmation).1% were excluded from the study due to improper patient selection (chief complaint or patient class outside the scope of study).Ease of useMAʼs received one hour of system training on the DARES tool.Patient encounters lasted between 7-12 minutes depending upon complexity.High marks for intuitive interface.Technical reliability - no software faults in 10 months of the studyCHAMP IIAssessed team triage in the most demanding environment (i.e. acute care encounters for new complaints) caring for the patient group most likely to require complex assessments.The enhanced care team requires the core knowledge engine be able to triage patient encountersIdentify those encounters that can be managed by the care team using established protocols.Identify outliers to protocols.Alert MA/RN to need for triage to a higher level of care.Determine urgency of referral to the next care level.Economic viability and utility of DARES in the clinic settingCosts and universal access issues evaluated by percentages of patients requiring assessments by midlevel providers or physicians.Throughput of the clinic using less expensive personnel must offset any added costs.ResultsTeam triage76% of patients did not require triage to higher level of care.No missed outliersResults compared to blinded physician assessments.24% of patients triggered a triage recommendation by the knowledge engine..Possible outlier.Potentially hyper-acute emergency.Need for confirmatory data outside of the MAʼs scope of practice.Complex multiple co-morbid conditions.Early triage of patients with findings associated with severe indolent diseases.
  • Article: 99 hypothetical case scenarios, and clinical data from 100 real patients, were used to test the performance of the Isabel pediatric DCS. The ‘correct’ or final diagnosis was known for all cases. Hypothetical cases were provided by 12 different pediatricians, and clinical data from real patients were collected from emergency departments at 4 NHS sites. Isabel suggested the ‘correct’ or final diagnosis in 91% of the hypothetical cases, and 95% of the real cases.A two-person expert panel also provided an ‘optimal’ set of 2-3 diagnoses, for each real case, that juniors would have needed to work up in order to ensure safe decision making. In 73% cases, Isabel displayed all such diagnoses.
  • Use data-mining to learn ethnicity-specific drug interactionsFor example, consider that Statins work differently in Indians and Japanese (Japanese need half the dose, Indian's might need a lot more .. in Indian people with baseline LDL >160 mg/dl and LDL-C goal of <70 mg/dl only 71-74 % will achieve goal even with the highest approved doses of atorvastatin and rosuvastatin.Data can help in figuring out if it's really worth it to saturate Indians with statins to reach the LDL goals learnt from a Caucasian population.practice-based evidence, learning new medical insights from routine care, a bit of a pun on evidence-based practice which all doctors understand and means "don't just trust your gut, rely on clinical trials". No new experiment needed, in a sense free virtual clinical trials, only limited by the amount of data (to get statistical power) and the number of signals (we can only answer questions about things we know about, if we only know about drugs and diagnostics we can't answer questions that include diet for example). Another thing I like to harp on is that clinical trials are a very limited way to create new knowledge: they are expensive (so we only ask when we suspect we know the answer), biased (find me 100 white guys 35-45 with only type 2 diabetes) and give no insight about what happens when one combines drugs or conditions. So they give some insight on what happens on a limited fraction of the axes in a very high dimensional space, but no insight of what happens in general in between these axes. In the US only about 30M people take regularly 3 or more drugs for 3 or more conditions, and about half of them are on 6 or more drugs. None of these combinations have ever been tested in a clinical trial and are unlikely to be. Studying after the fact could lead to safer cocktails, simply by substituting some of the drugs.Vioxx: using only Stanford data (less than 2M lives, biased toward serious conditions, more heart things than general practice) Nigam was able to show that the association between Vioxx and heart attacks was overwhelming (odds ratio with the entire confidence interval above 1) two years before the recall. He was able to show the same thing on 6 of the main 9 recalls of the past 10 years with the same advanced warning of 2 years.Drug interaction: many people take statins, and about 16,000 people per year in the US undergo an organ transplant (most commonly a kidney) and require an immunosuppressant. Two equally effective choices: Cyclosporin A and Tacrolimus. A doctor will prescribe at random. Data-mining by Nigam's team at BMI has shown that Tacrolimus is 3x safer, meaning a patient taking Cyclosporin A in combination with a statin has 3x the odds of rhabdomyolisis (serious muscle degeneration, they pee their muscles). It turns out this drug and all statins share an enzyme, so taking both at the same time is the same as overdosing on statins.Some context: drug-drug interactions account for 30% of all adverse drug reactions (ADR). ADR is the 4th leading cause of death in the US.Happy drug story: peripheral arterial disease (think of it as a heart attack in your leg, leading to limping and eventually amputation) has only one effective treatment: cilostazol. But it belongs to the same class as a drug shown in 1991 to cause heart attacks, so it carries a black label warning for patient with chronic heart failure of any kind which scares everyone, and is only prescribed 6% of the time (basically last resort). No study has ever implicated this drug directly though. Nigam showed that it is safe in all patients, even those with an existing heart condition. 
  • Example: Ayasdi and new types of breast cancer
  • Transcript

    • 1. 2025: 20% doctor included? an exercise in technology speculation & musings vinod khosla twitter: @vkhosla
    • 2. 10% to 20% of cases: delayed, missed, and incorrect diagnosis graber, et al., jama, 2005 2
    • 3. 40,000+ patients in u.s. icus may die with a misdiagnosis annually winters, et al., bmj quality & safety, 2012 3
    • 4. 50% of MDs are below-average math 4
    • 5. human doctors cognitive limitations cognitive biases 5
    • 6. a study of one hundred cases of diagnostic error involving internists found… graber, et al., jama, 20056
    • 7. …system-related factors contributed to the diagnostic error in 65% of the cases and cognitive factors in 74%... graber, et al., jama, 20057
    • 8. …premature closure was the single most common cause graber, et al., jama, 20058
    • 9. the value of second opinions|74|305|339|3429 cleveland clinic doctors’ review of initial diagnosis 22% 26% 18% 15% 11% recommend minor changes to treatment plan recommend moderate changes to treatment plan recommend major changes to treatment plan find need for further testing disagree with initial diagnosis
    • 10. the American College of Cardiology and the American Heart Association made 7,196 recommendations leading to 53 practice guidelines on 22 topics… …48% have level C evidence (the worst kind)… …11% have level A evidence (the best kind)… …and only 19% of recommendations in class I guidelines had level A evidence 10 tricoci, et al., jama, 2009
    • 11. surgeons were given detailed diagnoses & asked if patients should get surgery … half said yes … the other half said no … when asked again two years later, 40% of the docs gave a different answer eddy, et al., jama, 199011
    • 12. fifty-eight experts’ estimates of the chance of an outcome of an important procedure 0% 0.2% 0.5% 1% 1% 1% 1.5% 1.5% 2% 3% 3% 4% 5% 5% 5% 5% 5% 5% 5% 6% 6% 6% 8% 10% 10% 10% 10% 13% 13% 15% 15% 18% 20% 20% 20% 25% 25% 25% 30% 30% 40% 50% 50% 50% 62% 70% 73% 75 75% 75% 75% 80% 80% 80% 80% 80% 80% 100% what does a consensus of a group whose perceptions might vary from 0% to 100% even mean? eddy, et al., jama, 199012
    • 13. wide ranges of uncertainty eddy, et al., jama, 199013 seventeen experts’ estimates of the effect of screening on colon cancer deaths 0% 25% 50% 75% 100% proportion of colon cancer deaths prevented = one expert’s response
    • 14. conventional wisdom and the “tradition of medicine” matthews & wilson, new scientist, 201014 should fever be reduced in critically ill patients? “there were seven deaths in people getting standard treatment and only one in those allowed to have fever” “the team felt compelled to call a halt, feeling it would be unethical to allow any more patients to get standard treatment”
    • 15. nearly half of all american adults have difficulty understanding and acting upon health information institute of medicine of the national academies, 2004 15
    • 16. there is good reason to challenge the assumption that every individual practitioner's decision is necessarily correct eddy, et al., jama, 1990 16
    • 17. for most study designs and settings, it is more likely for a research claim to be false than true ioannidis, plos med, 2005 17
    • 18. entrepreneurs will ask the naïve questions that uncover hidden assumptions… …and move us to the grey zone of “speculations” 18
    • 19. in the future, patients will have the data & analysis to become the CEO of your own health peter diamandis 19
    • 20. 80% of what MDs do can be replaced (with better care than the average MD)… …but not every MD function will be replaced 20
    • 21. the “human” element of care can be provided by the most “humane” humans (and MDs can be humane) 21
    • 22. machines are better at integrative medicine… …across “all symptoms”, demeanor, patient history, phone activity, 1000s of data points, genomics, population management guidelines, … …and machines won’t have to win every time… …they’ll just be better overall 22
    • 23. Lifecom CHAMP in acute care I …distributed care with medical assistants were 91% accurate without labs, imaging, or exams II …“safe triage” with 75% physician bypass rate for acute care encounters 23
    • 24. isabel II matched expert diagnoses 91-95% of the time 24
    • 25. dr. algorithm v0 25
    • 26. the transition will start with “toddler MDs” and digital first-aid kits 26
    • 27. 27 Cellscope: ENT+ derm images… Adamant: breath analysis Eyenetra: auto-optometrist mental health Alivecor: frequent EKG+ analysis Quanttus: physiological metrics (HR, BP, SV, CO, RR, T, …) Medgle: graph of medicine Healthtap, Crowdmed: crowdsourced answers Kyron: practice based evidence Jawbone, Misfit: wellness wearables
    • 28. don’t wait days to take your daughter to the hospital… …check her ear infection as soon as it hurts 28 *a khosla ventures investment CellScope
    • 29. don’t go to the hospital and get connected to a bunch of electrodes… …take your own ecg for less than a buck… …and know you have heart disease before you have an attack! 29 *a khosla ventures investment AliveCor
    • 30. don’t go to the optometrist… …get measured for glasses at home 30 *a khosla ventures investment EyeNetra
    • 31. don’t guess what’s going on inside your body… …get vital intelligence 31 *a khosla ventures investment Quanttus
    • 32. don’t wait for an asthma attack… …know when it’s coming 32 *a khosla ventures investment Adamant
    • 33. forget kappas of 0.2 in the DSM-5… …get reliable, consistent diagnoses 33 *a khosla ventures investment
    • 34. graph the world of medicine… …and see where you fit 34 *a khosla ventures investment Medgle
    • 35. evidence-based medicine isn’t enough… …think practice-based evidence 35 *a khosla ventures investment Kyron … use data-mining to learn ethnicity-specific drug interactions (e.g. statins work differently in Indians)
    • 36. thousands of physicians… …no waiting room 36 *a khosla ventures investment HealthTap
    • 37. healthcare service stations & digital first aid kits 37
    • 38. keep people out of the doctor’s office… …with point innovations in cardiology, dermatology, optometry, psychi atry, internal medicine, … 38
    • 39. innocuous point innovations… 39 …will evolve into a wave and explode into a tsunami
    • 40. dr. algorithm v0 v1 – 2015 v2 – 2017 v3 – 2019 v4 – 2021 v5 – 2023 v6 – 2025 … 40
    • 41. we’ll start with clumsy point innovations like alivecor, cellscope, adamant,, neu rotrek, consumer physics, jawbone, misfit, … …“insighted” by machine learning… …leading us to discover things we never knew were right in front of us 41
    • 42. the best MDs will train systems over 10 years… …systems will symbiotically provide “bionic assist” and “AMPLIFY” MDs 42
    • 43. dr. house+++ will be the trainer for dr. algorithm …no manners required! …but manners learned! 43
    • 44. findings thanks to data using statins for in-hospital stroke patients reduced the death rate by 40%! kaiser permanente44
    • 45. 45 the practice of medicine the science of medicine
    • 46. I will be wrong on the specifics but directionally right 46
    • 47. the shift to “computerization” has already happened in other areas… …airline pilots, stock trading, car driving 47
    • 48. there aren’t enough rural doctors in india and few have access to jama journals, mris, … …the world of medicine is under-resourced globally 48
    • 49. 20% doctor included? vinod khosla twitter: @vkhosla