top of page

Deep Machine Learning vs YOU

Does the following data give you a sense of the exciting potential of 'machines' in predicting your future health better than you? Hopefully you read the previous blog on you predicting you the best.

Here is the reference:


We conducted a prospective population cohort study with a substantial sample size of 502,628 participants aged 40–69 years, recruited to the UK Biobank from 2006–2010 and followed-up until 2016. This comprehensive study assessed participants on a range of demographic, biometric, clinical and lifestyle factors, ensuring a robust and diverse dataset for analysis.

Mortality data by ICD-10 were obtained from linkage to Office of National Statistics. Models were developed using deep learning, random forest and Cox regression. Calibration was assessed by comparing observed to predicted risks; and discrimination by area under the ‘receiver operating curve’ (AUC).

Now, let's dive into the table of results. But before we do, let's break down the technical jargon for a clearer understanding.

  • Column 1 is the "predictor variable," a presumed consideration contributing towards longevity or early mortality.

  • Column 2 is the "hazard ratio" for dying young/prematurely

  • In column 2, a value higher than 1 infers a higher risk, while a value much lower than 1 infers a lower risk. Do not get hung up on the actual numbers - instead, focus on how far the value is compared to 1.

For example, for values less than one, a value of 0.5 means there are greater mortality changes for that "predictor variable" compared to one with a value of 0.8. (0.8 is closer to 1 than 0.5—thus, it reflects less risk.

Don't get too hung up on the actual values. Just realize that the closer a value is to "1," the less impactful it is on longevity—at least based on this machine-learning model. Heck, it should be spot-on if machine learning is any good because the database is half a million people!

In the chart below, I have put a red box around those that show the greatest variability—thus the greatest risk of dying young. One of the most interesting sets of values is ethnicity. Before I present the entire chart, here is the segment on ethnicity.

Thus, in the UK, at least, Whites die at a ~40% higher rate compared to Blacks, all else being equal. Ref for White is presumed 1.0, and for Black, 0.63 - which translates to a 37% lower death rate in that population.

Looking at the table below, machine learning figured out that a person's age is the biggest risk factor for dying soon!

It also figured out that if you have a serious disease, you are more likely to die soon.

Yes, I'm making fun of machine learning for the moment. I know it will improve. Garbage in - garbage out. However, can we call a half million people a "garbage" data set?

There are some gems in this data. For example, look at the very last row. Forced expiratory volume - a test you can do every day if you purchase a $20 device, is a superior predictor compared to: BP, poverty (deprivation index) BMI and MET (physical activity). Of course, the constellation of the risk factors leads to a low forced expiratory volume.

Can the machine tell you that?

Answer: NO

But a simple survey probably can.

The bottom line: If we continue to measure the same things, adding a bit more precision will not lead to much change.

Maybe I wrote this because I'm afraid of being replaced. For example, in France, e-bikes ridden by portly older women passed me and my daughter on our tandem regularly! But I think she still preferred to be with me.


Previous Blog on the value of listening to YOU about your health.

Getting labs can be exciting. Do I have inflammation? What else do I have? What about a genomic test? When you get the news, you can explain that your issues were not within your control! Remember, genetics load the gun, but your environment pulls the trigger (thank you, Dr. Stark, for this poignant explanation of that type of testing)

Neither of these types of tests answers the crucial questions How? and Why? At best, they answer the question "What?" But within your sense of "self" are the right answers.

The standard of care relies on answering the question "What?" so you can be put on a symptomatic (and often harmful) synthetic material that seems to make you feel better—at least for the short term. But we all want to enjoy a robust health span, right?

Who is your best doctor? Who can answer the key questions about your health? These questions are:

  1. Why is my health not optimal?

  2. How can I restore my health?

Your doctors—even your functional doctors—may not help you answer these questions if they rely on a myriad of tests that just answer "what?"


I've always been an advocate for health surveys. Yes, it's boring. It's not a scalar energy treatment or a med bed, or some other fancy treatment that has moved far away from what human health really is.

Surveys have their shortcomings. They can be impersonal, and people can "stretch the truth" due to fear that the truth will be used against them. However, in a world where you doctors look at a screen, but not you, at least this is an outlet to tell part of the story of your life and health.

I conducted a test of our health survey versus deep-dive labs. I had a group of 20 people run our chronic disease risk assessment (CDRA) and an extensive panel of labs. I then did a 1-hour health consult with each individual.

  • First, I ran down the entire survey and put together a plan based on their answers.

  • Then, I reviewed their labs using our science-based reference ranges.

Guess what? In all 20 cases, I almost made no changes to their plan. The labs added very little insight into "how" to restore/optimize their health. The exceptions were related to pathogen testing, which sometimes led to adding a specific treatment.


We demonize the standard of care, but even a stopped clock is right twice/day. Here is such an important instance. This article was published in the Journal of the American Heart Association!

"Participant‐Reported Health Status Predicts Cardiovascular and All‐Cause Mortality Independent of Established and Nontraditional Biomarkers: Evidence From a Representative US Sample"


Self‐rated health provides prognostic information beyond that captured by

  • traditional ASCVD risk assessments and by

  • nontraditional CVD biomarkers. (fibrinogen, C‐reactive protein [CRP])

  • Consideration of self‐rated health in combination with traditional risk factors may facilitate risk assessment and clinical care.


  • a 1 standard deviation decrease in self‐rated health was associated with increased risk of CVD mortality (hazard ratio [HR], 1.92; 95% CI, 1.51–2.45; P<0.001),



  • this hazard remained strong after adjusting for ASCVD risk and nontraditional biomarkers (HR, 1.79; 95% CI, 1.42–2.26; P<0.001).


  • Self‐rated health also predicted all‐cause mortality even after adjustment for ASCVD risk and nontraditional biomarkers (HR, 1.50; 95% CI, 1.35–1.66; P<0.001).


Getting biomarkers and other tests into optimal ranges is important, but .....


How do you obtain self-rated health?

  1. Listen

  2. Survey

  3. Probe (based on survey)

  4. Listen

You have all heart this quote and probably understand its meaning.

"The king is dead; long live the king!" is a traditional proclamation made following the accession of a new monarch in various countries. The seemingly contradictory phrase simultaneously announces the death of the previous monarch and asserts continuity by saluting the new monarch.

So surveying is DEAD (as is focusing on the person/patient) - LONG LIVE THE SURVEY!


Weekly Webinar Links: Join us for detailed health information - at no charge. All are welcome.

Monday at noon EST -

Wednesday at 8 pm EST -


Be Bold - Be Brave - Stay Well


125 views0 comments

Recent Posts

See All


bottom of page