Publications

Education, justification, and optimization are the cornerstones to enhancing the radiation safety of medical imaging. Education regarding the benefits and risks of imaging and the principles of radiation safety is required for all clinicians in order for them to be able to use imaging optimally. Empowering patients with knowledge of the benefits and risks of imaging will facilitate their meaningful participation in decisions related to their health care, which is necessary to achieve patient-centered care. Limiting the use of imaging to appropriate clinical indications can ensure that the benefits of imaging outweigh any potential risks. Finally, the continually expanding repertoire of techniques that allow high-quality imaging with lower radiation exposure should be used when available to achieve safer imaging. The implementation of these strategies in practice is necessary to achieve high-quality, patient-centered imaging and will require a shared effort and investment by all stakeholders, including physicians, patients, national scientific and educational organizations, politicians, and industry.

BACKGROUND

Guidelines recommend incorporating life expectancy (LE) into clinical decision-making for preventive interventions such as cancer screening. Previous research focused on mortality risk (e.g. 28% at 4 years) which is more difficult to interpret than LE (e.g. 7.3 years) for both patients and clinicians. Our objective was to utilize the Gompertz Law of Human Mortality which states that mortality risk doubles in a fixed time interval to transform the Lee mortality index into a LE calculator.

METHODS

We examined community-dwelling older adults age 50 and over enrolled in the nationally representative 1998 wave of the Health and Retirement Study or HRS (response rate 81%), dividing study respondents into development (n = 11701) and validation (n = 8009) cohorts. In the development cohort, we fit proportional hazards Gompertz survival functions for each of the risk groups defined by the Lee mortality index. We validated our LE estimates by comparing our predicted LE with observed survival in the HRS validation cohort and an external validation cohort from the 2004 wave of the English Longitudinal Study on Ageing or ELSA (n = 7042).

RESULTS

The ELSA cohort had a lower 8-year mortality risk (14%) compared to our HRS development (23%) and validation cohorts (25%). Our model had good discrimination in the validation cohorts (Harrell's c 0.78 in HRS and 0.80 in the ELSA). Our predicted LE's were similar to observed survival in the HRS validation cohort without evidence of miscalibration (Hosmer-Lemeshow, p = 0.2 at 8 years). However, our predicted LE's were longer than observed survival in the ELSA cohort with evidence of miscalibration (Hosmer-Lemeshow, p<0.001 at 8 years) reflecting the lower mortality rate in ELSA.

CONCLUSION

We transformed a previously validated mortality index into a LE calculator that incorporated patient-level risk factors. Our LE calculator may help clinicians determine which preventive interventions are most appropriate for older US adults.

BACKGROUND

Globally and in the United States, smoking and obesity are leading causes of death and disability. Reliable estimates of prevalence for these risk factors are often missing variables in public health surveillance programs. This may limit the capacity of public health surveillance to target interventions or to assess associations between other environmental risk factors (e.g., air pollution) and health because smoking and obesity are often important confounders.

OBJECTIVES

To generate prevalence estimates of smoking and obesity rates over small areas for the United States (i.e., at the ZIP code and census tract levels).

METHODS

We predicted smoking and obesity prevalence using a combined approach first using a lasso-based variable selection procedure followed by a two-level random effects regression with a Poisson link clustered on state and county. We used data from the Behavioral Risk Factor Surveillance System (BRFSS) from 1991 to 2010 to estimate the model. We used 10-fold cross-validated mean squared errors and the variance of the residuals to test our model. To downscale the estimates we combined the prediction equations with 1990 and 2000 U.S. Census data for each of the four five-year time periods in this time range at the ZIP code and census tract levels. Several sensitivity analyses were conducted using models that included only basic terms, that accounted for spatial autocorrelation, and used Generalized Linear Models that did not include random effects.

RESULTS

The two-level random effects model produced improved estimates compared to the fixed effects-only models. Estimates were particularly improved for the two-thirds of the conterminous U.S. where BRFSS data were available to estimate the county level random effects. We downscaled the smoking and obesity rate predictions to derive ZIP code and census tract estimates.

CONCLUSIONS

To our knowledge these smoking and obesity predictions are the first to be developed for the entire conterminous U.S. for census tracts and ZIP codes. Our estimates could have significant utility for public health surveillance.