Google Calendar is going to count down remaining life expectancy based on DNA methylation.
Well, maybe not, but it could be a news headline that you may see more of in the coming days.
Within less than a week after the publication of the scientific study, we already have "Scientists' 'hugely exciting' DNA research may lead to extending life", "DNA 'Clock' Could Predict How Long You'll Live" and "Is Your DNA Clock Running Fast? You May Die Sooner Than Later".
A new study, published last week in Genome Biology, found that people who are biologically older than their actual age, die earlier. Combining the data of four studies, of more than 5,000 participants in total, the study observed a 16% higher risk of mortality per 5 years difference between biological ('DNA') and actual age.
The headlines in the news, not surprisingly, are catchy, but seem to be 'hugely' exaggerated interpretations of the actual study. This time the authors had fueled the imagination of the journalists by an overly promising title of their article ("DNA methylation age of blood predicts all-cause mortality in later life") and by sending out an enthusiastic press release.
The main conclusion sounds stunning but is grossly overinterpreted and possibly not even be true.
First, the word 'predict' in the title and article is asking for confusion and misinterpretation, as demonstrated by the early media coverage. The word often has a different meaning in scientific studies, including this one.
For most people, 'predict' means that, in this example, DNA methylation can tell who will die earlier, but scientists often use the word instead of 'is associated with', namely in studies where the 'predictor variable' is measured earlier in time than the 'predicted variable'. Because DNA methylation was measured--obviously--before and independent of people's later death, they say that methylation predicts mortality.
The researchers also wrote that DNA methylation 'significantly predicts', which further suggests that the prediction was good. Yet, in scientific articles 'significantly' only refers to statistical significance, not at all to how good, relevant, useful or meaningful the prediction is. The statistical significance of an observed association (e.g., DNA methylation is associated with later mortality) depends on the strength of the association but even more by the size of the study. A risk increase from 10 to 11.6 percent or from 20 to 23.2 percent (which illustrates a 1.16-fold increase or 16% higher risk, similar in size to what was observed in the DNA methylation study) is easily statistically significant in a study as large as this one. Yet, whether such small increases in risk can meaningfully predict death, is a different question that was not evaluated.
Second, when demonstrating that DNA methylation predicts mortality, it is important that alternative explanations are ruled out. When there are factors that influence DNA methylation and also increase the risk of mortality, then these factors could be the reason why an association between DNA methylation and mortality is observed. For example, if cardiovascular disease leads to changes in DNA methylation and to earlier death, then that could be an alternative explanation for the association.
The researchers aimed to rule out the potential effect of age, sex, smoking, education, education, diabetes, high blood pressure and cardiovascular disease. In two of the four studies they additionally considered the role of cognitive ability at young age, APOE status and social class. In these two studies, the strength of association lowered from 1.17 to 1.08 and from 1.22 to 1.12. The other two studies, the Framingham Heart Offspring Study and the Normative Aging Study, did not have this information available and observed only a minimal change in association (from 1.37 to 1.40 and from 1.11 to 1.09). The question is what would have been the change in these associations if the information had been available? When a similar decrease in association can be expected, the overall association will likely drop from 1.16 to a 1.06 (read: 1.06-fold increased risk) and may not be statistically significant.
Third, and most remarkably, was this information really not available? A previous publication based on data of the Framingham Heart Offspring Study had investigated the relationship between APOE and carotid atherosclerosis and reported complete availability of APOE data in the 6th cycle of the study. This DNA methylation study was conducted in those who had survived up to the 8th cycle of the study. And also socioeconomic status has been investigated in both the Framingham Heart Offspring Study and the Normative Aging Study. The authors do not explain why these APOE and socioeconomic status were not available for their analysis, and this clearly is an omission in the article.
Aging is a process that changes the entire body and expectedly on a detailed molecular level too. Future research will identify novel biomarkers that indicate faster or slower aging on the basis of which predictive tests can be developed. But we are not there yet. And likely not even close.