A reader put me onto Emily Oster, an economist who had the idea to apply basic data analysis to prenatal and parental decision-making. “Data-driven parenting” is a good schtick, and her credibility as an Ivy-league economist seals the deal.
One might be inclined to say that “data-driven parenting” is just a balm for the parental anxieties of the numerate. Perhaps one should give up the illusion of control? But Oster’s shows how certain data is actionable. The key insight is that while the data is public, the decisions are personal.
That’s how empirical decision theory works; available data combines with personal benefits or costs to drive a decision.
For example, in her book Expecting Better, she discusses the decision to do prenatal screening for chromosomal abnormalities. She discusses how one’s decision to whether or not to bring a fetus with chromosomal abnormalities to term relies on their values and personal circumstances and the data. Statistics summarize the data into age-based baseline odds of abnormalities, precision and accuracy of non-invasive vs. invasive tests, risk of miscarriage for invasive tests, as well as empirical estimates of rates of miscarriage for invasive tests.
Here’s how the analysis broke down for my wife and me.
My wife is at high risk for chromosomal abnormalities, so we did a prenatal genetic screening test. These tests work by collecting cell-free fetal DNA in the mother’s bloodstream and subjecting them to genetic analysis.
Here is part of the results reporting numbers for Down syndrome:
That reads as 1/62 baseline “risk” (risco geral) based on my wife’s age and the number of weeks into the pregnancy, and an estimate of < 1/10000 personal risk (risco pessoal) post-screening. I assume “risk” here means frequentist probability, as in “Down syndrome occurs in 1/62 pregnancies for this group.”
That was a relief. By the way, cognitive scientists who are reading this, how do I quantify that relief? In information-theoretic terms assuming that our personal chances are right up against the upper bound of 1/10000, that test result constituted .02 bits of information learned. That’s not much, I guess information gain doesn’t capture relief. Perhaps “relief” is better captured by the information differential between our outcome and the would-be outcome where baseline risk went up (~6 bits). But I digress.
How believable are these results? Should we be confident that our chances of Down syndrome are below 1/10000?
I found a few meta-analyses (cited below) that give about 99.3% sensitivity for this type of screening. They are all influenced by a large study mentioned in Oster’s book that covered 146,958 women in China (Zhang et al. 2015). In that study, there were 726 cases of Down syndrome, and 6 (or .8%) of the cases went undetected. That same study had about a .05% false-positive rate, meaning that .05% of those women received results that indicated Down syndrome only to find out later that the fetus did not, in fact, have Down syndrome. Another study of high-risk women (like my wife) showed 98.6% (209/212) sensitivity, and a false-positive rate was 0.2% (3/1471).
It is essential to note the false positives also vary with age group. False positives are quite high for someone in their early 20’s (~ 55%) and lower progressively with age.
I remember Kobe memes about being a “girl-dad.” Is “data-dad” a thing?
Mackie, F.L., Hemming, K., Allen, S., Morris, R.K. and Kilby, M.D., 2017. The accuracy of cell‐free fetal DNA‐based non‐invasive prenatal testing in singleton pregnancies: a systematic review and bivariate meta‐analysis. BJOG: An International Journal of Obstetrics & Gynaecology, 124(1), pp.32-46.
Taylor-Phillips, S., Freeman, K., Geppert, J., Agbebiyi, A., Uthman, O.A., Madan, J., Clarke, A., Quenby, S. and Clarke, A., 2016. Accuracy of non-invasive prenatal testing using cell-free DNA for detection of Down, Edwards and Patau syndromes: a systematic review and meta-analysis. BMJ open, 6(1), p.e010002.
Zhang, H., Gao, Y., Jiang, F., Fu, M., Yuan, Y., Guo, Y., Zhu, Z., Lin, M., Liu, Q., Tian, Z. and Zhang, H., 2015. Non‐invasive prenatal testing for trisomies 21, 18 and 13: clinical experience from 146 958 pregnancies. Ultrasound in Obstetrics & Gynecology, 45(5), pp.530-538.