data&donuts
  • Data & Donuts (thinky thoughts)
  • COLLABORATor
  • Data talks, people mumble
  • Cancer: The Brand
  • Time to make the donuts...
  • donuts (quick nibbles)
  • Tools for writers and soon-to-be writers
  • datamonger.health
  • The "How" of Data Fluency

CANCER: THE BRAND

The asymptotic half-life of facts

6/18/2018

 
Picture
Do you remember calculating half-lives in high school chemistry? A single atom of a radioactive substance decays at an unpredictable rate. You might get lucky and witness the decay immediately or  It might take as many as 10000 lifetimes--pick a comfortable chair. 

​But in aggregate we can make estimates based on the law of large numbers. Similar to scientific facts--we can calculate the point where half of the atoms (or knowledge around a specific topic or discipline) has decayed or according to complexity scientist Samuel Arbesman, "...measure the amount of time for half of a subject's knowledge to be overturned." 

To understand the decay in the "truth" of a paper we can measure how long it takes for the citation of an average paper in a field to end. Whether it is no longer interesting no longer relevant or has been contradicted by new research this paper is no longer a part of the living scientific literature. It is out-of-date. The amount of time it takes for others to stop citing half of the literature in a field is also a half-life of sorts. 

Sam brilliantly distinguishes facts based on their rate of change. There are fast-changing facts like weather, slow-changing facts like the number of continents (think Pangea) and somewhere in the middle--mesofacts. Mesofacts change as well but typically along the scale somewhere within the human lifetime.

Read his book, The Half-Life of Facts: Why Everything We Know Has an Expiration Date
for more depth, it is truly fascinating. For example the half-life of physics is ~ 13 years while math is churning away around 9 years.

Why does this matter? Think of how accurate your calculation would be about the decay of a single atom of uranium. It is only when a large collection of atoms have accumulated that we can understand the rate of decay. The law of large numbers is quite simply a law of probabilities--or as the number of events or sample size grows--the difference between theoretical and actual approaches "0" or becomes closer to the average of a population.
In the field of Oncology--and beyond--these are critical points to remember. The problem with modern R&D and discovery is that we are heralding each unique uranium atom and betting on the outcome of the rate of decay. Early research is exploding in the media and investors are frenetic. What we know is incremental. Unless you work in science or clinical research you tend to hear the proclamations and the hype. The problem is each new hypothesis begins with an observation.

​It may be a fact today but what happens when replicated or expanded outside of clinically controlled environments? We once waited for data--for certainty or at least the probability of certainty. Now we want to rush discovery into "fact" status at the cost of accuracy or actual scientific methods. If you attended ASCO this year--you know what I mean. Investors want something to believe in and we are only too eager to provide fodder--even if it is quick to decay.

Here is a link to tweets on the problem with precision oncology. Vinay Prasad describes the cost of rushing to approve drugs where we lack sufficient facts to be certain that benefits outweigh risks.
Picture
If the topic fascinates you, here are two podcasts that fueled my curiosity. The Half Life of Facts and The Person You Become. 
    Bonny is a data enthusiast applying curated analysis and visualization to persistent tensions between health policy, economics, and clinical research in oncology.
    Follow @datamongerbonny

    Archives

    November 2020
    January 2020
    October 2019
    August 2019
    July 2019
    January 2019
    December 2018
    August 2018
    June 2018
    April 2018
    March 2018
    December 2017
    November 2017
    September 2017
    June 2017
    May 2017
    April 2017
    October 2016
    September 2016
    August 2016
    April 2016
    March 2016
    November 2015
    October 2015
    September 2015
    July 2015
    June 2015
    May 2015
    April 2015

    Categories

    All

    Subscribe to our mailing list

    * indicates required
    Email Format
Proudly powered by Weebly
  • Data & Donuts (thinky thoughts)
  • COLLABORATor
  • Data talks, people mumble
  • Cancer: The Brand
  • Time to make the donuts...
  • donuts (quick nibbles)
  • Tools for writers and soon-to-be writers
  • datamonger.health
  • The "How" of Data Fluency