Tag: statistics

Biological Research Often Incorrectly Reports Sex Differences

Photo by Tim Mossholder on Unsplash

An analysis of published studies from a range of biological specialties shows that when data are reported by sex, critical statistical analyses are often missing and the findings are likely to be reported in misleading ways.

The analysis was published in the journal eLife.

“We found that when researchers report that males and females respond differently to a manipulation such as a drug treatment, 70% of the time the researchers have not actually compared those responses statistically at all,” said senior author Donna Maney, a professor of neuroscience in Emory’s Department of Psychology. “In other words, an alarming percentage of claims of sex differences are not backed by sufficient evidence.”

In the articles lacking the proper evidence, she added, sex-specific effects were claimed almost 90% of the time. In contrast, authors that tested statistically for sex-specific effects only reported them 63% of the time.

”Our results suggest that researchers are predisposed to finding sex differences and that sex-specific effects are likely over-reported in the literature,” Prof Maney said.

The problem is so pervasive not even her own work was safe. “Once I realised how prevalent it is, I went back and checked my own published articles and there it was,” she said. “I myself have claimed a sex difference without comparing males and females statistically.”

Prof Maney stressed that the problem should not be discounted; it is becoming increasingly serious, she said, because of mounting pressure from funding agencies and journals to study both sexes, and interest from the medical community to develop sex-specific treatments.

Better training and oversight are needed to ensure scientific rigor in research on sex differences, the authors wrote: “We call upon funding agencies, journal editors and our colleagues to raise the bar when it comes to testing for and reporting sex differences.”

Historically, biomedical research has often included just one sex, usually biased toward males. In recent decades, laws have been passed requiring US medical research to include females in clinical trials and report the sex of human participants or animal subjects.

“If you’re trying to model anything relevant to a general population, you should include both sexes,” Prof Maney explained. “There are a lot of ways that animals can vary, and sex is one of them. Leaving out half of the population makes a study less rigorous.”

As more studies consider sex-based differences, Maney adds, it is important to ensure that the methods underlying their analyses are sound.

For the analysis, Prof Maney and co-author Yesenia Garcia-Sifuentes, PhD candidate, looked at 147 studies published in 2019 to see what is used for evidence of sex differences. The studies ranged across nine different biological disciplines, including field studies on giraffes and immune responses in humans.

The studies that were analysed all included both males and females and separated the data by sex. Garcia-Sifuentes and Prof Maney found that the sexes were compared, either statistically or by assertion, in 80% of the articles. Of those articles, sex differences were reported in 70% of them and of those treated as a major finding in about half.

Statistical errors were seen in some studies, with a significant difference for one sex but not the other counted as a difference between them.  The problem with that approach is that the statistical tests conducted on each sex can’t give “yes” or “no” answers about whether the treatment had an effect.

“Comparing the outcome of two independent tests is like comparing a ‘maybe so’ with an ‘I don’t know’ or ‘too soon to tell,'” Maney explains. “You’re just guessing. To show actual evidence that the response to treatment differed between females and males, you need to show statistically that the effect of treatment depended on sex. That is, to claim a ‘sex-specific’ effect, you must demonstrate that the effect in one sex was statistically different from the effect in the other.”

Conversely, their analysis also encountered strategies that could mask sex differences, such as pooling data from males and females without testing for a difference.

“At this moment in history, the stakes are high,” Maney says. “Misreported findings may affect health care decisions in dangerous ways. Particularly in cases where sex-based differences may be used to determine what treatment someone gets for a particular condition, we need to proceed cautiously. We need to hold ourselves to a very high standard when it comes to scientific rigor.”

Source: EurekAlert!

Another COVID-scale Pandemic in 59 Years ‘Statistically Likely’

Photo by Edwin Hooper on Unsplash

A new study based on 400 years of historical records asserts that extreme pandemic events such as COVID are more common than believed.

The Duke University study, published in Proceedings of the National Academy of Sciences, used records of past outbreaks to estimate the intensity of those events and the yearly probability of them recurring.

It found the probability of a pandemic with similar impact to COVID is about 2% in any year, meaning that someone born in the year 2000 by now would have about a 38% chance of experiencing one. That probability is only increasing, highlighting the need to adjust perceptions of pandemic risks and expectations for preparedness, the researchers said.

“The most important takeaway is that large pandemics like COVID and the Spanish flu are relatively likely,” said study co-author William Pan, PhD, associate professor of global environmental health at Duke. The understanding that pandemics are not so rare should raise the priority of future prevention and control efforts, he said.

The study employed new statistical methods to measure the scale and frequency of disease outbreaks for which there was no immediate medical intervention over the past four centuries. Their analysis, including deadly pathogens including plague, smallpox, cholera, typhus and novel influenza viruses, found pandemics occurred with great variability in the past. But they also identified patterns that allowed them to describe the probabilities of similar-scale events happening again.

In the case of a pandemic like the Spanish flu, which killed more than 30 million people between 1918 and 1920, the probability of a pandemic of similar magnitude occurring ranged from 0.3% to 1.9% per year over the time period studied. Taken together, it is statistically likely that such a massive pandemic would occur within the next 400 years.

However, the data also show that the risk of intense outbreaks is increasing rapidly. Based on the increasing rate at which novel pathogens such as SARS-CoV-2 have broken loose in human populations in the past 50 years, the study estimates that the probability of novel disease outbreaks will likely triple in the next few decades.

With this increased risk factor, the researchers estimate that a COVID-scale pandemic is likely within a span of 59 years (by the year 2090), a result they write is “much lower than intuitively expected.” Although not included in the paper, they also calculated the probability of a pandemic capable of eliminating all human life, finding it statistically likely within the next 12 000 years. 

That does not mean it will be 59 years before the next COVID-like pandemic, nor that the Spanish flu for another 300 years. Such events are equally probable in any year during the span, said Duke University Professor Gabriel Katul, another of the paper’s authors.

“When a 100-year flood occurs today, one may erroneously presume that one can afford to wait another 100 years before experiencing another such event. This impression is false. One can get another 100-year flood the next year,” explained Prof Katul.

Dr Pan noted that population growth, changes in food systems, environmental degradation and more frequent contact between humans and disease-harboring animals all may be significant factors for increasing frequency of pandemics. However, he stresses that the statistical techniques are not to explain the pandemics.

However, he hopes the study will spark deeper exploration of the factors that may be making devastating pandemics more likely – and how to counteract them.

“This points to the importance of early response to disease outbreaks and building capacity for pandemic surveillance at the local and global scales, as well as for setting a research agenda for understanding why large outbreaks are becoming more common,” Dr Pan said.

Source: Duke University