1. Predicting cancer 
  2. Systems of loosely coupled variables 
  3. The non systems use of Odds Ratios. 

Predicting cancer 
Recently a friend recommended a book, The China Study: The Most Comprehensive Study of Nutrition Ever Conducted and the Startling Implications for Diet, Weight Loss and Long-term Health, showing with extensive research and literature review that having animal protein as a regular part of your diet is a cause of cancer. Wrong. To me, this interesting collection of research studies shows that a diet with animal protein is a statistically significant factor influencing development of cancer, but with a moderate association. A statistically significant influence but a very, very partial influence.

I will not address The China Study directly. Instead, I will address the fact that there is a long list of factors which are statistically significant predictors of cancer. To focus on just one variable misses the point about the interdependent system of variables which influence the development of cancer. How does one think about that?

What is the long list of cancer predictors? Animal protein in diet, oncogenes, family history of cancer, environmental toxins, alcohol, age, obesity, smoking, lack of exercise, inflammation, Socioeconomic Status (SES), race, Adverse Childhood Experiences (ACE), recent life events, level of differentiation of emotional functioning of the individual and of one’s family systems, active connection with extended family.

Adverse Childhood Events (ACE) provides a comparison to animal protein in one’s diet.  Animal protein in diet is a statistically significant predictor of cancer. ACE too is a significant predictor of cancer. What do you make of that?  Both are significant predictors.  How does one interpret that?<

Systems thinking about these matters would say not to look at any of these significant predictors in isolation.  They all work together, in an interdependent system, to predict cancer.

Animal protein is a significant predictor probably, but it doesn’t work alone.

A word about individual predictors.  This long list of predictors is pretty typical when one is trying to predict something important in a living system.  Any one of these predictors by itself has a weak or moderate association with the criterion variable, like cancer.

Be suspicious in studies of human beings when the researcher overfocuses on one variable.  Better to assume that any human phenomenon will have many influences, and then deal with what few exceptions emerge.

One way to use systems thinking with this state of affairs is the following.  Any time you consider one of the predictors, think to yourself, “What has to go with this single predictor, in order to get an effective prediction?”

Systems of loosely coupled variables 
It appears to me that variables taken from living systems constitute systems of loosely coupled variables, a concept I picked up from software engineers and organizational development people.

A system of loosely coupled variables is weakly to moderately interdependent, including with whatever important variable one is trying to predict, like cancer or early mortality. So, I think that the long list of cancer predictors is such a system of loosely coupled variables. Weakly to moderately correlated with each other and with important aspects of functioning which one is trying to predict, like cancer. No one variable is a big predictor by itself.

Implications. Animal protein in diet is by itself a moderate predictor of cancer, with plenty of exceptions to that trend. Cancer depends on animal protein and the living system provided by the variables in that long list of predictors. If you had  a number of adverse childhood experiences, get animal protein out of your diet. If not, you might keep eating animal protein, depending on some of the other predictors in the list. If you had a number of ACE, stop eating animal protein, find better differentiated people to hang out with, exercise moderately, maintain active connections with your extended family, etc.

The non-systems use of Odds Ratios 
Odds Ratios (OR)  are a currently faddish or fashionable statistical test to calculate and report. Journalists love it; it’s simple to express. “Exposure to X makes it 2.5 times more likely that you will come down with Y in the next five years.” 2.5 is the Odds Ratio. Both researchers and journalists rarely report how one needs to think about such results.

Systems thinking will point out the importance of context. So too with Odds Ratio. What is the prevalence in the general population of Y, the variable you are trying to predict? An OR of 2.5 which occurs in a  sample in which the prevalence of Y is 1.1% must be interpreted differently from an OR of 2.5 that occurs in a sample where the prevalence of Y is 23%.

In the 1.1% sample, an OR of 2.5 increases your prediction to 2.75. Presence of X leads one to predict an occurrence of Y of 2.75% in the 1.1% general population sample.  Then in the 23% sample, an OR of 2.5 will lead one to infer a prevalence of 57.5% among people exposed to X in the 23% sample. Big difference. Researchers and journalists often do not report this impact of context.

The March 5 New York Times Science section has an article that illustrates the statistics problems I’ve pointed to.

There is now a blood test purporting to detect the presence of amyloid plaque in the brain, even when a person has no symptoms of dementia. It is hoped that such a test would be an early predictor of later development of Alzheimer’s Disease. Some researchers even say that this test should be the definition of Alzheimer’s Disease.

As usual, the test results are interpreted with simplistic causal reasoning. There are different versions of this. All seem to believe that a statistically significant result for an association of two variables means that the variables are related in some kind of simple causal manner.

Just because variable A is twice as likely to be associated with the occurrence of variable B does not mean that A causes B.

The China Study: The Most Comprehensive Study of Nutrition Ever Conducted and the Startling Implications for Diet, Weight Loss and Long-term Health, by T. Colin Campbell and Thomas M. Campbell, 2004.

“Relationship of childhood abuse and household dysfunction to many of the leading causes of death in adults. The Adverse Childhood Experiences (ACE) Study”, by Felitti VJ, Anda RF, Nordenberg D, Williamson DF, Spitz AM, Edwards V, Koss MP, Marks JS, 1998.

“The Association Between Adverse Childhood Experiences and Risk of Cancer in Adulthood: A Systematic Review of the Literature”, by Dawn M. Holman, MPH, Katie A. Ports, PhD,  Natasha D. Buchanan, PhD, Nikki A. Hawkins, PhD, Melissa T. Merrick, PhD, Marilyn Metzler, RN, MPH, and Katrina F. Trivers, PhD, MSPH, 2016.

“Loosely Coupled Systems: A Reconceptualization”, by J. Douglas Orton and Karl E. Weick, The Academy of Management Review Vol. 15, No. 2 (Apr., 1990), pp. 203-223.

https://www.nytimes.com/2024/03/04/health/alzheimers-amyloiddiagnosis.html, “Apparently healthy, yet diagnosed with Alzheimer’s.