How can we best predict how long someone will live?

If you had to guess how long a person will live, what would you want to know about them? In other words, what single piece of information best predicts longevity?

This was a central question addressed by renowned demographer Noreen Goldman, from Princeton University, in a recent lecture at UCLA – and her answers were fascinating. In a large-scale study in Taiwan of aging adults (ages 54+), Dr. Goldman and her team interviewed and carried out comprehensive medical assessments of over a thousand men and women (Cornman et al., 2014). In these assessments, they gathered a large amount of data about each person: demographic characteristics (e.g., age, sex, marital and socioeconomic status), medical history (e.g., smoking status, health conditions and diseases, family history), and standard medical markers of health (e.g., blood pressure, cholesterol, BMI).  They also gathered data on markers of health that are not typically collected during standard physical exams, but have been shown in recent research to be robust markers of health - such as levels of stress hormones (e.g., cortisol, norepinephrine) and markers of inflammation (e.g., IL-6). Then, six years later, Dr. Goldman’s team followed up to see which of these participants were still alive. The researchers then asked the question, of all the many pieces of data we gathered, which best predicted who was still living and who was not?

The first best predictor of mortality wasn’t so surprising: it was the person’s age. The older the person, the more likely they were to have passed away at the six-year follow-up. But what was astonishing was what turned out to be the second best predictor of mortality. Was it smoking status? Sex (since men tend to die earlier than women)? Medical history? In fact, it was none of these. After age, the best predictor of mortality was…wait for it... IL-6. IL-6 (interleukin-6) is a marker of inflammation – it’s a hormone secreted by immune cells in order to stimulate an immune response. Inflammation is a necessary component of a healthy immune response to infection or injury, but when inflammation becomes chronically elevated it can have detrimental effects.  A growing body of evidence has implicated chronic inflammation in a number of diseases, including cardiovascular disease, cancer, diabetes, arthritis, and even depression.

It is striking that levels of IL-6 - something that is not typically even measured in medical exams - was a more robust predictor of mortality than any of the standard medical measures – including blood pressure, cholesterol, and BMI. What are the implications of this for standard medical care? Should measures of IL-6 be added to routine physical exams? Dr. Goldman’s findings require replication by other teams and in other populations before we get too carried away, but if the findings hold true in other studies, the implications could call for a reevaluation of standard medical practices.

Another fascinating finding of the study arose from the researchers’ comparison of  interviewer, physician, and self-reports of health to find out which was the best predictor of mortality. The researchers first had each subject participate in an hour-long interview with a member of the research team, who asked them a number of questions about their mental health, feelings of stress, and functioning. The interviewers also had participants perform a few tasks such as chair stands and timed walking tasks to get an idea of their physical functioning. Then, study interviewers were asked to rate the participant’s health as either “excellent”, “very good”, “good”, “fair” or “poor”. On another day, the researchers had the physicians rate participants’ health on the same scale. Finally, participants were asked to rate their own health. Then, after the six year follow-up, the researchers compared the three ratings (interviewer, physician, and participant self-ratings), to see which rating best predicted how long a person lived.

What they found was that participants who the interviewers had rated as having “poor” health were 11.2 times more likely to have died at the six-year follow-up – in other words, the interviewer ratings strongly predicted mortality. As for the individuals’ self-ratings, participants who had rated themselves as having “poor” health were 6 times as likely to have died at the follow-up six years later. So, the self-ratings also significantly predicted mortality, but not as powerfully as the interviewer ratings. Finally – and most surprisingly – the physician ratings were the only ratings of the three that did not predict how long individuals would live (Todd & Goldman, 2013). This is striking – individuals were better at assessing their own health than were professionally trained physicians. Furthermore, interviewers who asked about a host of things not typically asked about in medical exams (mental health, feelings of stress) did better at predicting death than either self-ratings or doctors. Should physicians consider their patients the real experts when it comes to their health status? (Yes.)  Should they consider incorporating questions about mental health into standard physical exams? (Definitely.) Again, Dr. Goldman’s findings need to be replicated before definitive conclusions can be drawn, but her research brings up a plethora of fascinating questions and intriguing potential implications.

To watch a video of the full seminar, sponsored by UCLA’s California Center for Population Research, visit:


Cornman, J. C., Glei, D. A., Goldman, N., Chang, M., Lin, H., Chuang, Y., …. Weinstein, M. (2014). Cohort Profile: The Social Environment and Biomarkers of Aging Study (SEBAS) in Taiwan. International Journal of Epidemiology. Advance online publication. doi: 10:1093/ije/dyu179.

Todd, M. A. & Goldman, N. (2013). Do interviewer and physician health ratings predict mortality? Epidemiology, 24(6), 913-920. doi: 10.1097/EDE.0b013e3182a713a8

UncategorizedLarissa Dooley