Machine Learning Times
Machine Learning Times
AI Success Depends On How You Choose This One Number
 Originally published in Forbes, March 25, 2024. To do...
Elon Musk Predicts Artificial General Intelligence In 2 Years. Here’s Why That’s Hype
 Originally published in Forbes, April 10, 2024 When OpenAI’s...
Survey: Machine Learning Projects Still Routinely Fail to Deploy
 Originally published in KDnuggets. Eric Siegel highlights the chronic...
Three Best Practices for Unilever’s Global Analytics Initiatives
    This article from Morgan Vawter, Global Vice...

10 years ago
Predictive Analytics will Revolutionize Healthcare


For more case studies in predictive healthcare, see Predictive Analytics World for Healthcare, October 2015 in Boston.

This past week I attended the Predictive Analytics World Healthcare Conference in Boston.  This was the first healthcare focused event for Predictive Analytics World started by Eric Siegel.  Dell Software, Jointly Health, Predixion, and Baptist Health were represented at the event.  The main topics of the conference were: the current wasteful system of United States Healthcare, clinical decision support and big data. Roles for big data in healthcare are high cost patients, readmissions, triage, decompensation, adverse events, and treatment optimization.

The Journal of Health Affairs compared the United States to other developed countries in 1975 and 2005, in both studies the United States was dead last, 12 out of 12.  In 2005 the United States was not only last, but it also paid twice as much as other countries for that position.  Don Berwick, an administrator for CMS said on average between 21% and 47% of the dollars spent on patients in the United States has no value.  The reasons for this waste are outlined below.

  • Failure of care of process

    • Failure of care processes includes preventative treatments or therapies that were not provided to the patient, usually because no one would pay for it or lack of patient engagement or adherence. 32 million Americans use 3 or medicines daily and 75% of adults are non-adherent in one or many ways. The estimated impact of non-adherence is estimated to cost 100 billion annually


    • When people think of fee for service they usually blame providers saying that they are greedy and use extra treatments to pocket the extra cash from CMS. However, many times patients demand extra treatment and why not if it is not invasive and they are not paying for it?  Before ACO’s and other evidence based payment systems the main vehicle of revenue for providers was fee for service.  A common example of overtreatment is prostate cancer.  Prostate cancer is a type of cancer that grows very slowly and many times patients with prostate cancer will end up dying from another disease or accident before their prostate cancer would have killed them.  Despite this fact, many elderly men are still receiving screenings and treatment for prostate cancer when it may not be in their best interest.  My grandmother was 86 years old, living on her own and self-sufficient when she received knee replacement surgery, a few months later she developed dementia.   She is now 91 and needs around the clock care and has a memory span of about 3 seconds.  None of her providers discussed the added risk of dementia for elderly patients with surgery.  She wanted the surgery so she could better tend to her garden and Medicare paid for it.  No one questioned it.  My mother’s quality of life is also reduced because she has to care for my grandmother.  Medicare pays for surgeries for seniors that don’t make sense but it does not pay for long term care.  Seniors need long term care, not surgeries.

    Administrative Complexity

    • There are entire floors of hospitals filled with administrative employees just to deal with all the paperwork and coding for CMS and insurance companies. With ICD-10 there are now over 16,000 codes to track diseases; the system is getting more complex, not less complex.  ICD-10 is putting hospitals at risk for 500K a year in lost revenue and fines from CMS.  Patients use different insurance companies and within those companies there are many different plans.  The hospital administrative staff has to correctly determine what these plans will and will not pay for, how much they are willing to pay and make sure they fill out all forms correctly the first time or the forms will be sent back.

    Failure of Care Coordination

    • The use of HIE’s is on the rise, but there are many practices that do not belong to an HIE. A patient can have the same tests done multiple times at different facilities which is wasteful and sometimes dangerous. There is also a large number of different HIE’s that do not share information.  For example, if your provider is a part of the MassHIway and you go on vacation and see a doctor that is a part of LAHIE, the information will not be shared.

    To fix the issues above Don Berwick created the Triple Aim Approach, which includes, improving population health, the patient experience and controlled per capita approach.

    Improve Population Health

    • Use evidence supported decisions
    • Have viable incentives for providers to improve the health of their patients
    • 5% of the population represent 50% of healthcare costs and the prevalence of chronic disease is rising, make sure to target these patients

    Improve Patient Experience

    • Change medical education to be more about how to find and use data and less about memorizing as much as possible. Usually the challenge for providers isn’t getting data, but how to use that data to help their patients.

    Controlled per Capita Cost

    • Use bundled payments instead of fee for service
    • Move costly inpatient procedures to outpatient if possible
    • Use copayments as incentives for patients to shop around for their procedures and treatments
    • Do not overtreat the elderly with procedures that may increase their lifespan a few months, but will greatly reduce their quality of life.

Healthcare Costs can be reduced 1 trillion dollars with Clinical Decision Support.  However, first we have to get providers to use clinical decision support.  Some providers currently have CDS available to them, but do not use it because they don’t have time, the data is statistically correct but not valuable, or the providers are overconfident.  To be effective, clinical decision support needs to fit into the provider’s workflow, be easy to understand, be accepted by fellow clinicians, hold up to scrutiny and be accurate and effective.  Clean data is also important, however if you wait for clean data in healthcare, you will get nothing done, you have to learn to work with uncertainty.
The current solutions for CDS are:

Clinical rules based solutions

  • This includes most vendor based solutions, which basically corresponds to medical school 101. For example, alerting the provider that the patient is at risk for a heart attack because they have high cholesterol and blood pressure.   These types of solutions cause fatigue for physicians,beacuse while these solutions are statistically correct, they are not useful.  They don’t give the provider any new information but create extra clicks.  Doctors hate nothing more than extra clicks.

Statistical algorithms

  • Not patient or disease specific
  • Katrina Belt of Baptist Health said they use LACE scores: Length of stay, Acuity, Comorbidity, Emergency room visits; often add Polypharmacy

Deep machine learning

  • Continually refined and more accurate as new data

Patient clusters

  • Combined capability to understand what isn’t inherently visible with clinical intelligence to more accurately identify patients at risk
  • map and predict patient-level risk

The 5 P’s of clinical support are


  • Be specific to the patient
  • For example determine which anti-depressant to use for a particular patient


  • The ACA is currently trying to address this with insurance for everyone including preventative screenings at no cost to the patient. The theory behind this is to get patients treated for symptoms that will if untreated land the patient in the hospital for an expensive inpatient visit.  However, because of these screening many more people will be tested and treated for diseases or ailments that they would have otherwise would not have been tested for.  Some of these tests will also return false positives which could lead to unneeded treatment or overtreatment
  • Prevent patients from getting a disease
  • Prevent patients who already have a disease from getting worse


  • Predict who will get the disease
  • Better prediction means better treatment, better treatment leads to better outcomes and fewer wasted resources


  • Look at patterns instead of random points of data


  • Using clinical support goes against tradition of provider’s relying on their memories
  • Assure providers that it is not decision making, it is decision support


  • If clinical decision support is widely used in healthcare, we have succeeded

The first Healthcare focused Predictive Analytics World conference was a huge success and I am grateful to have attended.  Predictive analytics in combination with current EHR’s can revolutionize medicine if done correctly.

By: Jamie Titak
Originally published at

One thought on “Predictive Analytics will Revolutionize Healthcare

  1. Pingback: Predictive Analytics will Revolutionize Healthcare | HENRY KOTULA

Leave a Reply