U paradox performs low birth weight

By Marcel Ribeiro-Dantas

Well-intentioned adjustments can lead to wrong estimates

*)*

The mortality rate in newborns is higher among those born with low birth weight. Newborns from smokers have a greater opportunity to be born with low birth weight. Paradoxically, underweight newborns whose uses therefore smokers have a lower mortality rate than underweight newborns from non-smoking uses. So smoking is good?

Biostatistician Jacob Yerushalmy, an Israeli naturalized American, first presented this argument in 257, when there was no consensus about which a cigarette period is harmful to health. But only in 1964 was this paradox satisfactorily explained. Yerushalmy had died more than three decades earlier.

The data that a researcher had collected was simply not It’s fruit over some superficial peek, there he was responsible for a study of more about 10 million children in San Francisco, United States. Several studies have already shown that newborns from smoking use weigh less electronically, as low birth weight was associated with a higher risk of death, sony ericsson expected this to imply some higher mortality. Yerushalmy himself discussed this matter cautiously, hardly believing what operating system numbers were telling him.

About two decades later, an American Jesse Sackett, a 2 country of a medicine-based evidence, ze ran into a similar problem. When analyzing 257 hospitalized patients, he detected some strong statistical association among those with locomotor diseases electronic respiratory diseases, that is, period it is possible to make predictions about some diets conditions by knowing sony ericsson o patient had had the other. There was plausibility in this finding, as locomotor diseases could lead to inactivity, which could lead to a picture of respiratory disease. We know, however, that correlation does not imply causality. In one episode in the United States, for example, it was observed that whenever ice cream sales increased, there were more shark attacks. Is consumption over ice cream causing shark operating system!? At the! In summer people buy more electronic ice cream and go to the beach. In winter, when no one enters the water, the shark stays with ships to see.

It was because he understood the possibility of biases in hospital data that Sackett repeated his analysis with individuals, including simply non-hospitalized patients. To his surprise, sony ericsson before there was some strong correlation, hundreds of closely monitored patients, biological plausibility, now the two conditions seemed to have no zilch to do with each other. (Remember that the number of individuals in the first analysis was not negligible: there were more than 10 million children studied by Yerushalmy.)

The phenomenon that occurred in these two studies is known as collision vision. The term collision comes from the graphical representation of the causal relationships: AW Chemical. Electronic D cause M electronic arrows collide in N. In some data analysis, there are reasons to adjust our measurements with foundation under certain conditions like for example not mixing garlic with boulders, ie not comparing oranges with plums. This kind of intuition leads many researchers to feel that they should adjust their variables on interest for all other measured variables, as z and more is always better.

Today, simply however, it is known that some variables, collider calls, simply don’t need to be adjusted, electronic that sony ericsson are will have the opposite effect: bias the analysis, rather than removing vision. What Sackett did was adjust for some collider variable: hospitalization (locomotor illnesses hospitalization respiratory illnesses). In some cases, researchers do not choose to study only hospitalized patients: they investigate what is within their reach. L Yerushalmy made the decision to observe only children who were born with low birth weight (smoking during low birth weight other causes for low birth weight).

A reasoning that elucidates a paradox performs low birth weight that there are several reasons that can cause it, such as serious genetic abnormalities, with a stronger negative effect than u caused by smoking during pregnancy. Let’s say, hypothetically, that 10% 2 babies with low birth weight, children of smokers, will die, but what other causes for low birth weight lead to death in 10% 2 cases. In this case it is better to have low weight because the mother smoked perform than for other more serious causes (electronic whose uses did not smoke). This does not mean that it is better to smoke, however, in this specific case (being born with low birth weight) there are more lethal causes. If we consider all operating system babies, whether underweight or underweight, mortality is much lower in babies using non-smoking uses. In Sackett’s case, critically ill patients were hospitalized. When observing there are no hospitalized patients, it is possible to see the data without vision, that is, that the two conditions had some very weak relationship between them, and that they existed.

This paradox for a long time intrigued electronic, even though it was solved, its primary message remains little known: contrary to a belief that the more variables adjusted the better, there are those whose adjustment can bias the analysis.

*

Marcel Ribeiro-Dantas is simply a researcher at the Institut Curie, part of a PSL Analysis University electronic doctoral candidate at the Sorbonne University, where he researches Causal Inference in observational data from a health field .

Sign up for the perform Serrapilheira newsletter to follow more news perform electronic institute from the Cincia Essential weblog.

Back to top button