Life Expectancy Trends Among Integrated Health Care System Enrollees, 2014–2017

Introduction: The Centers for Disease Control and Prevention (CDC) has reported downward trends in life expectancy and racial/ethnic differences between 2014 and 2017. Objective: To determine the life expectancy of the Kaiser Permanente Mid-Atlantic States (KPMAS) insured population as compared to the CDC National Vital Statistics data from 2014 to 2017. We also aimed to highlight the utilization of membership data to inform population statistical estimates such as life expectancy. We examine whether national trends in life expectancy are reflected in an insured population with relatively uniform access to care. Methods: This retrospective, data only study examined life expectancy between 2014 and 2017. Data from electronic medical records and the National Death Index were combined to construct complete life tables by race and sex for the KPMAS population, which was compared to the CDC National Vital Statistics data. Results: From 2014 to 2017, the overall KPMAS population life expectancy at birth varied between 84.6 and 85.2 years compared to the CDC reported national average of 78.6-78.9 years (p < 0.001). While the CDC dataset reported a 3.5- to 3.7-year life expectancy gap between non-Hispanic White and non-Hispanic Black populations, in the KPMAS population, this gap was significantly smaller (0.0-0.9 years). The gap in life expectancy between males and females was consistent across KPMAS and the CDC data; however, overall KPMAS male and female patient life expectancy was extended in comparison. Conclusion: Among members who disclosed their race/ethnicity, KPMAS Hispanic, non-Hispanic Black, and non-Hispanic White members had significantly higher life expectancies than the CDC dataset in all years reported.


Accounting for Non-Reporters
To examine the impact of non-reported race/ethnicity information on our study, we perform racial imputation using three different methodologies. In the first methodology, we impute only those members for whom we had no recorded racial data, and we combine this information with our known racial data to create a composite life expectancy with the best available information for all members. In addition, we impute race only for patients for whom there is recorded racial data; this allows us to estimate the quality of our imputation and the effect this imputation may have on the Non-Reporter population. Finally, we impute racial categorizations for all members of our population. This final imputation methodology helps us to incorporate information from all members equally.

Evaluation of Race/Ethnicity Imputation
To evaluate the imputation methodology, we impute data for those patients that have already identified their race/ethnicity. By and large, we find that the GEMS data is relatively good at classifying our members. The method tends to overweight the probability of a member being Non-Hispanic White and to underweight the probability of a member being an Asian or Pacific Islander. To demonstrate this tendency, we present a table for 2015 members showing the average GEMS probability of members coming from each race, grouped by members' reported race. We can also construct this kind of table for patient deaths. When we do so, we find similar overall tendencies; however, small departures can also lead to significant difference in life expectancy between algorithms.

Supplementary
In addition to examining the raw accuracy of this imputation, we examine how the methodology influences our conclusions. For most races, we find that imputation does not significantly alter our interpretation; however, imputation leads to a significant decrease in the estimated life expectancy of the Hispanic population. This effect arises because of the tendency of the partial allocation imputation method to cause mean-reversion. Because the Hispanic population has a high life expectancy, this effect significantly deflates their estimated life expectancy. The effect is muted in the Asian and Pacific Islander population because members outside of this population are unlikely to be misclassified into it.
This discrepancy indicates that we should treat imputation with a certain level of scrutiny, especially when using it to estimate the life expectancy of the KPMAS Hispanic population.

Effects of Racial Imputation
After performing all three imputation methods, we found that the overall effect of this data was marginal on all populations. We present this series of excess life plots below for completeness; however, the differences between these methods were not statistically significant in most cases.
For Hispanic members, imputation methods which resample the population of known Hispanic members (the green and orange lines) led to large underestimates of life expectation. In addition, inclusion of nonreporters had a slight impact on the population of Non-Hispanic White members (approximately 0.5 to 1 years of life expectation at birth, depending on the imputation methodology and year). This occurs because a significant proportion of non-reporters are estimated to be Non-Hispanic White.
We present the effects of the various imputation methods on our estimates of excess life expectation for the Hispanic, Non-Hispanic White, and Non-Hispanic Black populations below. Overall, we find that inclusion of Non-Reporters through imputation leads to very modest decreases in life expectation (the red and green lines); however, we also see that resampling only our known population (the orange line) can lead to significant problems, especially for the Hispanic population. Because of this instability, we decided to present the known population as our primary estimation methodology.
Despite these complications, conclusions regarding the life expectation of KPMAS members from these racial groups remain largely unaffected by the inclusion or exclusion of non-reporters. Unless we inaccurately resample our known population, inclusion of non-reporters only creates marginal effects that do not alter the primary conclusions of this study.

Supplementary Figure 1
Life Expectancy at Birth, by Year, Asian and Pacific Islander Population The Asian and Pacific Islander population at KPMAS experienced very high life expectancy during the study period. Life expectancy in this population ranged from 89.8 to 91.1 years during the study period, very similar to our Hispanic population. Unfortunately, the CDC does not publish statistics on the national average life expectancy for Asian and Pacific Islanders, so we cannot compare the KPMAS member population to the national average.

Supplementary Figure 2
Excess Life Expectancy In Tables 2 and 3, present a detailed look at average life expectancy at birth among members of KPMAS  and the CDC. Here we present Supplementary Tables 3 and 4, which demonstrate the Excess Life Expectancy of KPMAS members over the US National Average, grouped by sex, race, and year.