This web calculator of human mortality is based on the paper How Well Do the Standard Body-Mass Index or Variations With A Different Exponent Predict Human Lifespan? by Dean Foster, Howard Karloff, and Kenneth E. Shirley, to appear in the journal Obesity . The authors fit a Cox proportional hazards model to roughly 400,000 respondents from the NIH-AARP Diet and Health Study , a survey given in 1996-1997, in which times until death were measured through the end of 2009. The main goal of the paper was to better understand the relationship between BMI (both the traditional version and a new version described in the paper) and mortality, while accounting for the effects of several other health-related and demographic variables.

#### How to use this calculator

The calculator above takes the inputs provided in the dropdown menus and computes the estimated distribution of time until death for the given individual based on the statistical model that was fit in the above-referenced paper (Model M3, to be specific). The top plot shows the estimated distribution of the individual's lifespan, and the bottom plot shows the individual's relative risk of death as a function of his or her BMI, where BMI is defined in the traditional way, as one's weight in kilograms divided by the square of his or her height in meters (or, equivalently, as 703*(weight in lbs.)/(height in inches)^2.). The model is not causal, however, which means that changing the value of one variable would not necessarily cause a change in one given person's expected lifespan; rather, when one changes the values of one or more input variables, one should think about the results as comparisons between different groups of people.

#### A note on relative risk

The baseline relative risk of 1.0 is computed not from any particular 'baseline' individual, but rather is the average risk of death across all respondents measured at their observed BMI.

#### Potential sources of bias

The results from this data analysis are susceptible to three forms of bias:

1. Selection Bias: The original 3.5 million surveys were sent (in 1996-1997) to members of the AARP (American Association for Retired Persons); if members of the AARP at this time were different from the rest of the population, e.g., more healthy, then the results of the analysis would not generalize.
2. Non-response bias: Only about 566,000 people out of 3.5 million (16%) people who received the survey responded, and the response was voluntary. If the group of respondents were different from the group of those who received the survey (i.e., perhaps only comparatively healthy AARP members responded), then the results would be less generalizable.
3. Discarding chronically ill respondents: For the main analysis, the authors discarded data from respondents who were chronically ill, which was defined to be either sick from or diagnosed with cancer, heart disease, renal disease, emphysema, or having had a stroke. This resulted in removing 25% of the remaining respondents, many of whom were, presumably, less healthy than the respondents without a chronic illness.

The authors suspect that these three sources of bias explain the fact that the expected lifetimes produced by the calculator are approximately five years longer than those reported in the most recent CDC life tables (from 2011).

#### Source code

The authors cannot publicly share the data which were used in the analysis, but the code used to do the analysis is available on Github at https://github.com/kshirley/BMI . Please open an issue on the github repo if you have any questions or comments.

<insert life table here>