BEWARE, PERSON-YEARS! EXPERIENCE OF SIMPSON PARADOX OBSERVATION IN EPIDEMIOLOGICAL RISK EXAMINATIONS

Analiz Riska Zdorovʹû. 2017;(4):23-31 DOI 10.21668/health.risk/2017.4.02.eng

 

Journal Homepage

Journal Title: Analiz Riska Zdorovʹû

ISSN: 2308-1155 (Print); 2308-1163 (Online)

Publisher: FBSI “Federal Scientific Center for Medical and Preventive Health Risk Management Technologies”

Society/Institution: FBSI “Federal Scientific Center for Medical and Preventive Health Risk Management Technologies”

LCC Subject Category: Medicine

Country of publisher: Russian Federation

Language of fulltext: Russian, English

Full-text formats available: PDF

 

AUTHORS

V.F. Obesnyuk (Southern Urals Institute for Biophysics of the Russian Federal Medical-Biological Agency)

EDITORIAL INFORMATION

Double blind peer review

Editorial Board

Instructions for authors

Time From Submission to Publication: 20 weeks

 

Abstract | Full Text

It is shown, on the examples of concrete publications, that "person-years" category application in multi-factor health risks analysis can lead to false conclusions in the process of observation data grouping due to Simpson paradox influence when examinations are performed via demographic or epidemiological techniques. The paradox occurs when heterogeneous strata are being compared. "Person-years" category first appeared in the middle of the 17th century, long before first applications of mathematical tools in statistics and probability theory; it does not fully correspond to up-to-date requirements of epidemiological research. Risk theory should change 17–18 century paradigm as it focuses on conditional probability of unwanted events occurrence and not on a principle of comparing their intensities. It is particularly vital in case when we deal with determining possible damage to health caused by effects exerted by such factors and under such conditions when individual damage cannot be measured objectively but when it is possible to quantitatively determine regularities of changes in stochastic ability to survive for a large group of people or remote consequences occurrence for it. We prove it is necessary to create specialized mathematical tools and hybrid software able to solve a risks assessment task as an inverse one. Mathematical tools of large contingency tables could serve as prototypes of such tools; we can also use multi-factor logistical and Poisson regressions which are usually applied in countable events analysis. We should note that it is also necessary to eliminate a number of methodological drawbacks which are attributable to the said tools.