How does it all end? Trends and disparities in health at the end of life.

OBJECTIVES: To consider trends and disparities in end-of-life health in the US.
METHODS: I use data from the National Health Interview Survey, linked to death records through 2015, for respondents who died at ages 65+ to compare the prevalence of three health outcomes in the last six years of life across time, sex, age, race, and educational attainment. Self-rated health (SRH) is available for respondents interviewed in years 1987-2014, while information on activities of daily living (ADL) and instrumental activities of daily living (IADL) is available for the period 1997-2014.
RESULTS: By the end of the study period, individuals reported two fewer months of fair/poor health at the end of life than those dying in earlier years. In contrast, time lived with at least one activity limitation at the end of life generally remained comparable. Compared to men, women on average reported an additional year of living with an IADL limitation before death, and an additional eight months with an ADL limitation. Despite sex differences in disability, both sexes reported similar periods of fair/poor SRH before death. Similarly, while individuals who lived to older ages experienced a longer disabled period before death than individuals who died at younger ages, all age groups were equally likely to report fair/poor SRH. Black adults and adults with less formal schooling also spent more time with an end-of-life disability. For men, these racial and socioeconomic disparities lessened as death approached. For women, inequalities persisted until death. DISCUSSION: These findings suggest that despite increasing life expectancy, the period of poor health and disability prior to death has not recently been extended. Black women and women with less than a high school degree, require extended support at the end of life.

Introduction

The period preceding death has become an important and distinct stage of the contemporary life course [1,2]. Where death was once sudden, the sweeping health innovations of the last 150 years mean that death in the United States today often follows an extended period of chronic illness [1]. Although each death is influenced by a unique combination of social, behavioral, and genetic factors, there are also commonalities across this final stage. For many adults, disability and depression increase at the end of life, while cognitive skills and a sense of well-being begin to wane [3-6]. Typically, self-rated health, vision, handgrip strength, and weight also decline toward the end of life [4,5,7,8].

Rising life expectancy at older ages has raised concerns that the period of poor health and disability prior to death is growing. Research typically addresses this topic with the implicit assumption that advancing age is the main risk factor for declining health. However, the onset of several health conditions, including end-of-life depression and cognitive decline, is more closely linked to years of life remaining than years lived [4,7,8]. Comparing the health of older adults who are the same proximity to death (for example, comparing all adults in their last year of life) may yield different insights than comparing adults who are the same age, but differing distances from death (for example, comparing all 70 year-olds).

In this paper, I examine trends and inequalities in aging from the perspective of time to death, rather than time since birth. I compare three indictors of health—self-rated health (SRH) and two self-reports of disability—in the last 6 years of life among adults dying at ages 65+ across time, sex, age, race, and educational attainment. SRH is a subjective and self-reported indicator of health. While the two disability measures are also self-reported, they serve as more objective assesments of requiring assistance. This study is the first to examine annual trends in SRH at the end of life, as well as the first to produce national estimates of end-of-life SRH for several subpopulations. Quantifying end-of-life processes is crucial to both the success of programs aiming to meet the needs of a growing older population and to empower individuals to create advanced care plans about the end-of-life care they wish to receive.

Background

Years-to-death

From evaluating the financial wellbeing of pension systems to predicting a population’s healthcare needs, the end-of-life period is of interest across disciplines. A significant analytic decision in this research is how to measure time. While most research considers the time elapsed since birth, some approaches measure backward from the other end of the lifespan: death. Years to death can be a proxy for the complex and interacting social, behavioral, environmental, and genetic processes that determine each individual’s moment of death. The usefulness of a variable for remaining lifetime was first described in the 1970’s (see Sanderson & Scherbov [9] for a history of the variable, as well as a demonstration of using the variable to study population aging). An allure of the variable is that its utility remains under-explored, despite yielding new perspectives that can be missed if using only chronological age.

An important analytic decision when using the variable is the maximum length of the retrospective period before death. While some of the studies cited throughout this paper consider the last one or two years of life [10-12], others extend 3–8 years before death [6,13-16], and some well beyond 10 years [4,7,8]. Lunney et al.’s [15] finding that racial disparities in disability are “erased” in the last 1–1.5 years of life suggests that a period longer than 2 years before death is needed to capture evolving patterns of disparities.

Gerstorf et al. [4] find that well-being among older adults in several countries begins to decline 3–5 years before death, around the same time as cognitive abilities [6]. Stenholm et al. [8] find that, compared to similarly-aged respondents who did not die, deceased participants of the Health and Retirement Study had a higher prevalence of poor SRH as early as 11–12 years before death. In a small sample of males ages 60+, Alley et al. [7] document that weight loss typically begins as early as nine years prior to death. Raab et al. [16] examine tandem trajectories of mental health and disability in the last eight years of life, while Gill et al. [10] document five disability trajectories in the last year of life. Although most individuals in Gill et. al’s sample were not disabled 12 months before death, more than half were severely disabled in the last month.

Trends in healthy aging

As individuals age, many develop at least one chronic condition. One approach for estimating the impact of morbidity on day-to-day functioning is to determine whether an individual has difficulty performing Activities of Daily Living (ADLs) or Instrumental Activities of Daily Living (IADLs). ADL’s include basic tasks such as dressing and eating, while IADL’s encompass activities that facilitate independent living, like grocery shopping or balancing a checkbook. ADL limitations are strong indicators of requiring physical assistance, with roughly 40% of community-dwelling adults age 65+ with one limitation and nearly 90% with 3+ receiving caregiving help [17]. While IADL limitations are less disabling than ADL limitations, an IADL limitation indicates that an individual requires some level of support in order to live independently.

An influential extraneous force shaping trends in disability prevalence is the changing composition of the population. Given the sweeping changes of the twentieth century, younger cohorts are reaching older ages having had better childhood health and more educated parents, reduced exposure to physically-demanding jobs, and higher levels of educational attainment—all factors linked to postponed age at onset of limitations [18]. More recent cohorts of older men are also less likely than their predecessors to be heavy smokers [19]. Because cohorts are evolving at the same time as the contexts in which they live, it is difficult to separate period and cohort effects. Crimmins et al. [20] speculate that once-disabling conditions may be less disabling today due to factors such as earlier diagnosis and better disease management, improved housing environments, and technological changes.

The age-specific prevalence of some disabilities may be declining over time, though findings are sensitive to analytic choices. Although the age-specific prevalence of ADL limitations declined in the 1990’s [18,21,22], evidence on whether more recent cohorts are less likely to experience IADL limitations is conflicting [18,22]. Other work using broader measures of disability suggests that recent increases in life expectancy at age 65 were primarily driven by increases in disability-free years [23].

Research on trends in health and disability by years of life remaining, rather than years lived, is limited. In a working paper, Cutler et al. [14] find that the prevalence of limitations in the last 5 years of life declined by up to 14% in the early 1990’s, but that the trend remained flat the following decade. This latter finding is echoed by Smith et al. [12], who find no trend in the prevalence of disability in the last two years of life for decedents dying between 1995–2010. In contrast, Beltrán-Sánchez et al. [13] find that the cohort of people dying in the late 2000’s reported a higher prevalence of chronic conditions in the final six years of life than did the cohort dying between 1998–2004. The increase in chronic conditions at the end of life may be a recent phenomenon, as Cutler et al. [14], using slightly older data, find no significant change throughout the 1990’s in the prevalence of major chronic conditions in the last three years of life.

Inequalities in healthy aging

The processes translating health inequalities at younger ages into inequalities at older ages are nuanced. On one hand, inequalities in health may be magnified with age. Adverse health experiences might accumulate over the life course and interact with vulnerabilities that accompany old age. The implications of socially-patterned health behaviors from younger years could also be postponed to older ages, such as the lag between cigarette smoking and the onset of lung cancer. On the other hand, not even the most privileged groups are exempt from aging—a fact which may level health inequalities as age advances [24,25]. Another possibility through which inequalities may diminish with age is selective mortality. Since some populations are exposed to systematically higher mortality rates throughout their lives, these groups can be highly select by the time they reach older ages. By nature of their design, studies that use chronological age (comparing 80 year-old White adults to their 80 year-old Black peers, for example) ignore the influence of selective mortality. While the issue is greatly lessened when considering time-to-death (comparing racial differences five years before death, for example), it nevertheless persists anytime a study sample has a minimum age below which differential mortality occurs.

Generally, conclusions about how health inequalities evolve across the life course depend on whether time is measured by elapsed age or proximity to death. Although older Black adults experience a higher prevalence of disability compared to their White peers of the same age [15,26,27], Lunney et al. [15] find that Black-White differences in disability are “erased in the final 1 to 1.6 years before death”, suggesting that “dying eliminate[s] a health disparity.” This result is consistent with work finding no racial differences in disability in the two years before death [12]. Some racial disparities, however, persist. Raab et al. [16] find that Black decedents are more likely than their White peers to exhibit adverse combinations of disability and poor mental health in the last eight years of life. Warner and Brown [27] test for several possible explanations, including the mediating roles of adult socio-economic status, health behaviors, and marital status. The authors find that while these variables account for Black-White differences in limitations for men, they do not fully explain the disadvantage for Black women. In a study similar to the present one, Liao et al. [11] find while Black-White differences in outcomes like long-term disability and hospital stays are persistent over the last two years of life, they are mostly explained by differences in educational attainment. In a review of research on end-of-life quality, Carr & Luth [1] examine how significant racial and socio-economic disparities in creating end-of-life plans may influence disparities in quality of life before death.

Just as certain age-specific racial disparities in health, disparities by educational attainment in health remain at older ages. Educational differences in the number of healthy years an average individual could expect to live in the 1980’s and 1990’s were even larger than differences in overall life expectancy [28] and Gini coefficients for health inequalities by education have been shown to increase with age [29]. Educational disparities in the age of onset of ADL limitations have also widened since the 1990’s [30]. Unlike the leveling effect impending death has on some racial disparities in disability by time-to-death, socio-economic differences appear to persist into the final stage of life. High school dropouts are more likely to be disabled in the last two years of life than high school graduates [12], and individuals with a terminal high school degree or less are more likely than those with at least some college education to report a combination of disability and poor mental health at the end of life [16].

Methods

Data

I use data from the 1987–2014 National Health Interview Survey (NHIS), downloaded from IPUMS [31]. The NHIS, conducted annually by the National Center for Health Statistics (NCHS), is a cross-sectional health survey of the civilian, non-institutionalized U.S. population that has been linked to death records through the end of 2015. While residents of long-term care institutions are not included in the baseline sample, the sample may include individuals who were interviewed at home and then moved to an institution during the follow-up period. More information on the survey is available on the IPUMS website (nhis.ipums.org).

The NHIS began consistently asking respondents about ADL and IADL limitations in 1997, but has included an annual question on SRH since the 1970’s. Since respondent matching to death records did not begin until 1986, however, and because a different weighting scheme was used in 1986, I begin the analysis of SRH trends in 1987. The analysis of ADL and IADL measures begins in 1997. I examine all outcomes until 2014, the last year for which interviewed respondents in the public-use files have been linked to death records (through 2015).

The sample

I restrict the sample to respondents who died at age 65 or above, within 6 years of being interviewed. I estimate the prevalence of each health outcome for all respondents not missing information on the outcome variable. The SRH analysis sample consists of 77,295 individuals across all years 1987–2014. Sample sizes for the IADL and ADL analyses, which span years 1997–2014, are 40,354 and 40,359, respectively.

I assign each decedent a value for years to death by subtracting the calendar year in which a respondent was interviewed from the respondent’s year of death reported in the linked mortality file.

I consider the last six years of life. Since the annual trends portion of the analysis requires six years to have elapsed since interview and mortality data is only available through 2015, the last year for which estimates can be produced is for respondents interviewed in 2008. Although a window longer than six years would be optimal, it strikes the balance between observing outcomes and disparities for as long as possible while tracking relatively recent trends. The categorical variable for years to death ranges from 1 (decedents died within 0–11.9 months of being interviewed) to 6 (decedents died within 5–5.9 years of interview).

Self-rated health is a predictor of subsequent mortality (see Jylhä [32] for why this may be). The NHIS asks respondents to categorize their health as excellent, very good, good, fair, or poor. For some of the analyses, I dichotomize answers into a dummy variable for unfavorable health (fair or poor SHR).

The NHIS asks questions on six different ADLs: whether a respondent requires help eating, bathing, dressing, moving about the home, using the toilet, or getting in/out of bed. Consistent with existing research, I classify an individual as having a disability if s/he requires help in performing at least one of these activities [12,22]. In the Supplemental Materials, I provide more fine-grained results for individuals with 1, 2, or 3+ ADLs.

Though IADL limitations are less disabling than ADL limitations, their presence indicates that an individual requires some support in order to live independently. The NHIS ascertained information on IADL limitations in the NHIS using a single yes/no question for whether a respondent needed help for “handling routine needs, such as everyday household chores, doing necessary business, shopping, or getting around for other purposes.” While the above ADL variable combines six survey questions, the IADL variable reflects only this single question.

Population characteristics

I consider patterns in end-of-life health across three socially-stratifying characteristics that are linked to differential health outcomes across the life course: sex, race (non-Hispanic Black, non-Hispanic White), and educational attainment (

Analytic approach

I estimate the proportion of the population reporting a given health outcome at each year x before death. This proportion can be interpreted as the proportion of total person-years lived with the given health outcome in year x before death—or as the average proportion of year x each individual spends in the health state. A 0.4 annual prevalence of poor SRH, for example, indicates that 40% of the population reported poor health, that 40% of all person-years lived were in poor health, and that, on average, each individual spent 40% of the year, or approximately 5 months, in poor health. To estimate the total number of the final six years that are spent in poor health, I conceptualize respondents as belonging to a synthetic cohort assembled along years to death and sum the prevalence across each year to death.

To compare outcomes across population subgroups, I estimate the mean of an outcome prevalence across the study period (years 1997–2014 for SRH; years 1997–2014 for disability).

For the time trends analysis, I stratify results across two-year interview periods. Since respondents interviewed in later years were not exposed for the full six years, I limit the sample to respondents interviewed before 2009 for this portion of the analysis. To estimate 95% confidence intervals for the sums of averages, I use bootstrapping procedures to repeat the estimation 500 times and take the 2.5th and 97.5th centiles.

All analyses were performed using Stata version 16 (Statacorp), account for the complex survey design of the NHIS using the svy package, and employ the NCHS-recommended weights (mortwt in IPUMS). Data and code for replication are available on the author’s webpage.

Results

Table 1 shows the nationally-representative distribution of population characteristics and the mean time between interview and death for each subgroup. The table also reports the median ages at death (median instead of mean, as age is top-coded). Of those who survive to age 65, most adults also survive to age 80, with women outliving men by three years (median ages at death of 83.2 vs. 80.3). Among both sexes, non-Hispanic Black individuals tend to die three years earlier than their White peers. Notably, women with less than a high school education in this population die at older ages than college-educated women (median ages of death of 84.1 vs. 82.7). Since levels of educational attainment among U.S. women increased considerably over time, women who are oldest at interview—and therefore older at death—are disproportionately less educated, which results in their paradoxically older median age at death (the reverse is true at younger ages). The age at death has climbed since the beginning of the survey, with those interviewed in 2013–2014 dying about 1.5 years older than those interviewed 15 years earlier.

Table 1

Characteristics of adults 65+ dying within 6 years of NHIS interview, 1986–2014.

Standard deviation in parentheses.

Characteristic	Females (N = 20,639)b			Males (N = 19,745)b
	%	Median age at deathc	Mean yrs to deathd	%	Median age at deathc	Mean yrs to deathd
Overall	--	83.17	3.53 (0.01)	--	80.25	3.42 (0.01)
Age at death
65–74	22.04 (0.36)	70.75	3.43 (0.03)	30.29 (0.37)	70.25	3.34 (0.02)
75–84	35.93 (0.42)	80.58	3.47 (0.02)	39.67 (0.37)	80.25	3.43 (0.02)
85+	42.03 (0.47)	85.00	3.63 (0.02)	30.04 (0.40)	85.00	3.49 (0.02)
Race e
Non-Hisp. White	89.24 (0.32)	83.67	3.54 (0.01)	90.87 (0.28)	80.67	3.41 (0.02)
Non-Hisp. Black	10.76 (0.32)	80.33	3.46 (0.03)	9.13 (0.28)	77.08	3.46 (0.04)
Educational attainment
<High school (HS)	33.10 (0.43)	84.08	3.51 (0.02)	30.94 (0.40)	80.67	3.39 (0.02)
HS/Some college	56.77 (0.42)	82.50	3.57 (0.02)	50.76 (0.40)	79.67	3.43 (0.02)
BA or more	10.13 (0.28)	82.67	3.48 (0.04)	18.30 (0.37)	80.83	3.45 (0.03)
Year of interview
1997–98	12.97 (0.28)	82.25	3.69 (0.03)	12.40 (0.26)	79.50	3.62 (0.04)
1999–00	12.32 (0.26)	82.17	3.71 (0.03)	12.59 (0.27)	79.50	3.59 (0.03)
2001–02	12.08 (0.28)	82.83	3.69 (0.03)	12.79 (0.26)	80.17	3.53 (0.03)
2003–04	12.63 (0.28)	83.42	3.67 (0.04)	12.31 (0.29)	80.75	3.57 (0.04)
2005–06	12.61 (0.27)	83.67	3.78 (0.04)	11.99 (0.27)	80.58	3.67 (0.04)
2007–08	12.20 (0.34)	83.67	3.67 (0.04)	11.99 (0.34)	80.42	3.57 (0.04)
2009–10	12.48 (0.34)	83.58	3.62 (0.04)	12.69 (0.31)	80.42	3.49 (0.04)
2011–12	8.55 (0.26)	83.75	2.73 (0.03)	8.66 (0.22)	80.00	2.78 (0.03)
2013–14	4.16 (0.16)	83.83	1.73 (0.03)	4.58 (0.18)	80.83	1.68 (0.03)

a. All percentages and proportions weighted using IPUMS mortality weights mortwt.

b. Sample sizes in the header reflect the number of respondents in the surveys who died within 6 years of being interviewed, at age 65 or above. The SRH analysis begins in 1987, but for brevity, details are not included in this table. Each set of analyses is additionally restricted to respondents not missing information on the given health outcome. 20,582 females and 19,694 males were not missing information on self-rated health. 20,622 females and 19,732 males were not missing information on ADL limitations. 20,630 females and 19,729 males were not missing information on IADL limitations.

c. Median, rather than mean, age reported since age in the NHIS is top-coded at 85. For this reason, the median age at death for 85+ year-olds is 85.0.

d. The NHIS/IPUMS public-use files include respondents’ quarter of death and month of interview. By assuming that respondents were interviewed on the 15th of the month and died half-way though the quarter, I add some precision to the years-to-death variable.

e. Due to small sample sizes, non-White and non-Black respondents are excluded from race-specific analyses (but included in all other estimates).

On average, individuals died about 3.5 years after being interviewed. The mean time to death is shorter for decedents interviewed between 2010 and 2014 simply due to the design of this analysis (the post-interview exposure period is less than six years).

Fig 1 shows time trends by sex in the average time out of the last 6 years of life spent in fair/poor health or with limitations. The accompanying S1 Appendix provides more detailed estimates, with numeric results for each level of SRH and number of limitations. Men and women were equally likely to report adverse health, with both sexes on average spending 2.5–2.8 years of the last 6 in fair/poor health. Though levels fluctuate across interview years, the general trend is toward a slight decline in time spent in adverse health since the 1980’s. Compared to participants interviewed in 1987–88, respondents interviewed in 2007–08 spent roughly 2 months less in fair/poor health.

Fig 1

Years out of last six years of life spent in given health state, over time (with 95% confidence intervals).

Fig 1 also shows results for time spent with at least one IADL or ADL limitation for decedents interviewed between 1997–2008. Although men and women were equally likely to report adverse health at the end of life, women reported either kinds of limitations for much longer. Men spent just over 1 year requiring help with at least one IADL task, and roughly 8 months with at least one ADL limitation (compared to over 2 years for IADL and 13 months for ADL help among women). While there was no change for men in these measures over the study period, trends are noisier for women. Time spent with either type of limitation increased by two months between the most recent interview periods, disrupting the slight trend toward declining IADL limitations in previous years. S1 Appendix, which divides ADL limitations into 1, 2, and 3+ limitations, suggests that fluctuations in women’s time spent with at least 1 ADL limitation were driven by changes in time spent with 2+ limitations.

Figs 2–4 pool data across interview years 1997–2014 and plot the mean proportion of each year spent in each health state, by sex and population characteristic. Dashed lines connect the values for each prevalence to reflect the trajectories experienced by synthetic cohorts. S2 Appendix complements these figures with numeric values and more detailed health outcomes, as well as with results for men and women overall.

Fig 2

Proportion of population in given health state across final years of life, by age at death (with 95% confidence intervals).

Fig 4

Proportion of population in given health state across final years of life, by educational attainment (with 95% confidence intervals).

The first of these figures, Fig 2, compares decedents who die at ages 65–74, 75–84, and 85+. Regardless of age at death, the prevalence of poor health and disability increase in the years preceding death, more than doubling for most outcomes. The trajectory of worsening SRH is remarkably similar across ages at death, with about half of the last year of life lived in adverse health (also interpretable as half of the population reporting adverse SRH in the last year). Regardless of age, most adults are similarly disabled six years before death (with the exception of older women requiring more IADL assistance than younger women). However, in contrast to SRH, a distinct age pattern emerges as death draws closer. While disability trajectories look similar for those dying between 65–74 and 75–84 years-old, the decline is much steeper for the oldest men and women. In their final year of life, more than 45% of women who survive to age 85 require help with at least one basic care task such as bathing or walking, compared to ~30% of women who die at earlier ages. Over half of the age difference in time spent with an ADL limitation in Fig 2 is due to older decedents being more likely to have 3+ limitations (S2 Appendix). Nevertheless, older decedents report being in better health for slightly longer. While the oldest women spend 8 and 14 months longer with an IADL and IADL limitation than the average of their younger counterparts, they report 4 more months of good-to-excellent health. Similarly, the oldest men experience 4–6 more disabled months than younger men, but 5 months more of favorable health (S2 Appendix).

Fig 3 compares trajectories for decedents who identify as non-Hispanic White and non-Hispanic Black. Black-White differences in the last six years of life are most apparent among women, especially in the proportion reporting fair/poor health. Over half of Black women report being in fair/poor health six years before death, compared to a third of White women. Of all population subgroups listed in S2 Appendix, Black women spend the longest time in fair/poor health (3.44 years), nearly 1 year longer than White women (2.47). Even though Black women die at younger ages, ages which are typically associated with shorter periods of disability, Black women also spend the most time with severe disability. On average, Black women require help with at least three basic care tasks, like eating or bathing, for a full year before their death (compared to 7 months for White women). These differences persist into the last year of life: 70% of Black women report fair/poor health one year before death, compared to 55% of White women. Racial differences in total time spent in adverse health are about half the size for men and decrease to statistical non-significance in the last year of life.

Fig 3

Proportion of population in given health state across final years of life, by race (with 95% confidence intervals).

Fig 4 shows results by educational attainment. Educational gradients in end-of-life SRH are more pronounced than differences in disability. Even six years before death, the three education groups (

Discussion

Despite concerns about expanding morbidity at the end of life, I find that the amount of time individuals report unfavorable health in the last six years of life declined two months from 1987–2008. To the author’s knowledge, this is the first study to examine trends in SRH at the end of life. I also find no change in the length of time spent with at least one end-of-life IADL or ADL limitation from 1997–2008, barring a slight increase in the most recent period. These findings are generally consistent with work by Smith et al. [12], who document an unchanging prevalence of ADL limitations from 1995–2009 in the last 2 years of life. While Cutler et al. [14] use repeated cross sections of the Medicare Current Beneficiaries Survey linked to death records from 1991–2009 to document a decline in the prevalence of ADL and IADL limitations in the last 5 years of life in the 1990’s, they find no significant change in the following decade (the main focus of this analysis). My findings are in contrast to those by Beltrán-Sánchez et al. [13], who consider six major chronic conditions and find that the adult disease burden may have grown over a similar period. Perhaps our findings differ because of the operationalization of disability versus chronic conditions. A growing disease burden might not translate to a higher prevalence of reporting one or more limitation, especially if the increases in chronic conditions are among people who already have at least one disability.

I find that the presence of disability does not always translate into a perception of poor health. Although respondents who reach ages 85+ are more likely to be disabled at the end of life than those who die before age 85 (a finding similar to the one described by Smith et al. [12]), older decedents are slightly more likely to report being in good health in their final years. Stenholm et al. [8] find a similar pattern in the Health and Retirement Survey. One reason for this paradox may be that a perception of health is constructed by comparisons with reference to individuals of a similar age [32].

Another illustration of the disconnect between disability and perceived health at the end of life is sex differences therein. It is well-documented that women report worse health than men at the same ages because women are more likely to develop disabling conditions [33]. Even at the end of life, women in the present analysis spend twice as long as men living with an IADL limitation and 70% more time with an ADL limitation in the final six years. Surprisingly, sex differences in SRH at the end of life do not reflect the expected pattern. Both sexes are equally likely to report fair/poor health at the end of life, even though women report considerably more disability. The mechanisms through which death levels sex differences in SRH, but not disability, warrant further investigation.

In contrast to the small differences in SRH but sizeable differences in disability by age, the opposite is true for racial and educational gradients. Especially among men, racial and educational differences in disability are relatively small six years before death, while differences in SRH are sizable. As men approach death, these gradients become smaller or even converge in the final 1–2 years of life, consistent with findings by Lunney et al. [15]. While disparities also disappear for women in Lunney et. al’s study, Black women and women with lower levels of formal education in the present study consistently spend more time in fair/poor health and with a disability, even at the very end of life. Lunney et al.’s study population has a different age structure and is limited to adults living in Memphis and Pittsburgh. Perhaps the end-of-life experiences of women living in these two cities are not entirely representative of patterns among women nation-wide. While past work suggests that racial differences in some outcomes are largely explained by racial differences in educational attainment and socio-economic status [11,27], the explanation remains unsatisfactory for women [27]. Future work should consider other, traditionally non-measured factors, like stress and discrimination [34,35], as well as the intersection of racism, sexism, and ageism.

The implications of health decline at the end of life extend to other generations. Informal caregiving provided by relatives is the most common form of elder care [17]. Black individuals are more likely to be caregivers for family members, to spend more time caregiving, and to care for a family member with 3+ ADL limitations [36]. These facts are consistent with the findings of the present analysis. Black adults (particularly women) not only require care for longer, but require more intensive care. I find that the majority of racial disparities in one or more ADL limitations is driven by adults reporting three or more disabilities (S2 Appendix). Future work should examine which chronic conditions drive these racial disparities in higher order disability. Furthermore, given the unequal distribution of the financial, physical, and emotional burden of prolonged caregiving [37], future research should also consider the extent to which informal caregiving of family members is a vehicle for the intergenerational transmission of inequality.

Limitations

The NHIS does not interview individuals who at the time of interview live in a long-term care facility, though respondents who enter a care facility at some point after their NHIS interview are still linked to their death certificate. In other words, a non-institutionalized respondent can be interviewed by the NHIS before moving into long-term care some time later and remain in the follow-up group. The effects of excluding institutionalized individuals from the baseline interview are likely minimized by the typically short duration of residence in end-of-life care facilities [23,38]. The median length of stay for nursing home residents is 5 months, and 53% of residents die within 6 months of admission. The median stay is longer for women (8 vs. 3 months for men), Black individuals (6.5 vs. 5 months for Whites), and poorer groups (9 months for the bottom income quartile vs. 3 for the top) [38]. Since the sickest members of these groups are more likely to be institutionalized at baseline, and therefore excluded from the sample, between-group estimates are likely conservative. Evidence of convergence in disparities in end-of-life health in this analysis should therefore be interpreted with caution.

Another bias to consider when comparing differences across groups is selective mortality. Since the analysis only includes adults who survive to age 65, the estimates of between-group differences are likely smaller than they would be if mortality before age 65 were random.

A limitation of repeated cross-sectional data is that it is not possible to distinguish between period and cohort effects. Measuring the relative importance of cohort composition versus period changes in the treatment of illness and disability would help target relevant interventions. For example, knowing that older male cohorts are less likely than their predecessors to be heavy smokers [19], but more likely to face the health problems associated with obesity [39], could inform the decision to divert funds away from tobacco-related interventions and toward new programs targeting obesity.

Conclusion

I report three main findings. First, despite rising ages at death, the findings indicate that the period of poor health and disability prior to death has not been extended in recent years. In this analysis, time in unfavorable health in the last six years of life declined by two months from 1987–2008, and time spent with at least one activity limitation from 1997–2008 remained stable. Second, even though older decedents and women are disabled for longer at the end of life, they report similar health to that of younger decedents and males, respectively. This paradox stands in contrast to well-studied sex differences at older chronological ages at which women report worse health and more disability. Third, the cross-sectional data suggests that while death reduces or even equalizes all racial and educational disparities among men, inequalities in healthy aging at the end of life persist for women.

Previous work finds that unequal access to formal education has a significant influence on end-of-life inequalities [11,27] and adds another item to the long list of benefits to expanding educational access. Minimizing disparities in educational outcomes is a long-term approach to reducing disparities at the end of life. In the meantime, the symptoms of health inequality could be addressed with additional support for those who report longer periods of ill health before death. In particular, women in this analysis require 6–12 more months of help with a limitation than men (for a total of 14 months lived with an ADL and 26 months with an IADL limitation). Similarly, Black women and women with less than a high school education require assistance for longer and report prolonged periods of unfavorable health. Ensuring access to and knowledge about programs that pay family members, including spouses, to act as a caregiver in affected communities may be beneficial. Support groups, including telephone hotlines, might be sources of social support for aging adults and caregivers. Finally, given the significant racial differences in reporting multiple limitations, future work should explore disparities in specific chronic conditions to create targeted interventions.

Years out of last six years of life spent in each health state for decedents 65+, over time.

(DOCX)

Click here for additional data file.

Years out of last six years of life spent in each health state for decedents 65+, by age at death, race, and educational attainment.

(DOCX)

Click here for additional data file.

25 Jun 2021

PONE-D-21-16309

How does it all end?

Trends and disparities in health at the end of life

PLOS ONE

Dear Dr. Vierboom,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The two reviewers, both experts in the areas of health, aging, and end of life research, have given careful consideration to your manuscript.  Please address each of their thoughtful comments in your response letter.  Both reviewers would like to see a better contextualization of the study in the existing literature, and more justification of some of your analytic decisions, for example: Why 6 years? Why these cause of death groupings?  In addition, please address further the important limitation of the NHIS exclusion of the institutionalized population, and the selection effect that presents.  Please explain what is meant in the Limitations section where it says that the newly-institutionalized NHIS respondents "remain in the sample" since the NHIS is cross-sectional.  Also, please explain. the decision to exclude Hispanics and limit comparisons to non-Hispanic whites and Blacks.

Please submit your revised manuscript by Aug 09 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see:  http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at  https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Ellen L. Idler

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

Additional Editor Comments (if provided):

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: I indicate above that the statistical analysis has been preformed appropriately and rigorously. However, I raise concerns about how certain variables are operationalized and indicate some analysis that might be missing. Based on the analysis plan reported by the authors (and I indicate is lacking), analysis appears to have been performed appropriately and rigorously. Please see attached file for additional comments.

Reviewer #2: This well-written descriptive paper provides a detailed statistical portrait of health and disability at the end of life, with careful attention to differences therein on the basis of sociodemographics (age, sex, race, education) and cause of death. The analyses are carefully done and use excellent NHIS data. Despite these strengths, the author could do more to justify their study aims and their key analytic decisions. I also encourage the author to say more about the possible influences of age vs. cohort effects, and selective survival when interpreting their results. I hope these comments are helpful to the author as they further develop this important project.

1. Why does the analysis focus on the final six years of life? Please provide a brief rationale for this decision. Other time points, such as last year of life, may align more closely with the literature on end-of-life medical expenditures, for instance. This shorter observation would help the author to better contextualize this in the literature, and link their findings to topics like expenditures. I suspect that the six-year decision was driven by sample size, but there may be other more compelling motivators.

2. More generally, the author could make a much more compelling case for the study goals. Why is this descriptive analysis helpful? What does it tell us, and how can these results advance research and policy/practice on end-of-life care?

3. PLOS is for a general readership, so more context is needed for readers who are not specialists in end-of-life topics. I would suggest a brief discussion of the strengths and weaknesses of using time-to-death measures to characterize end-of-life, and the relative strengths and weaknesses relative to measures like proxy reports and expenditure data that are other ways to characterize the end-of-life experience.

4. The section on inequalities in healthy aging, again, could do more to convey the importance and value of the research. This might be an opportune place to discuss the possibility of selective survival, such that lower SES and Blacks who survive until age 65 may show some health advantages relatives to higher SES and Whites who survive. This might be another interpretation of the seemingly counter-intuitive results whereby the least educated group appear to fare better than their more educated counterparts.

5. I would suggest using a more nuanced measure of age in the analysis, such as 65-74, 75-84, 85+. The current two category measure is quite coarse and some nuanced patterns may be concealed.

6. A minor issue. On page 13, the phrase “perceived amount of time individuals spend in unfavorable health…” is misleading. Study participants were not administered perceived life span or perceived probability of survival questions. Please re-word so that the sentence more accurately characterizes the data.

7. The analyses should raise issues of age versus cohort effects (and even touch on period effects) more directly, in the background and discussion. Even though the time period is fairly narrow, the youngest participants in 2014 and the oldest participants in 1987 represent vastly different cohorts, who have had different exposures to medications, public health interventions, assistive devices, etc. over the life course. At the very least, the limitations could say more about the inability to discern age vs cohort effects using repeated cross-sectional data.

Best of luck with your revision.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: Manuscript PONE-D-21-16309.docx

Click here for additional data file.

16 Dec 2021

Dear Professor Idler and Reviewers,

Thank you for your thorough examination of my manuscript. Your thoughtful feedback has helped me draft what I think is a stronger and more precise manuscript.

I outline the biggest changes to the analysis and the text below, before listing and responding to each reviewer’s comments. My responses in the Word version of this response document are written in blue and different font. To clearly distinguish my responses in the online version of these comments, I also use AR (for “author’s response”) before my reply to each comment. Revisions to the manuscript itself are recorded in track changes in the revised manuscript file.

Significant changes to the analysis:

1. Cause-of-death (COD) analysis: I have removed the cause-of-death analysis, in part due to Reviewer 1’s concerns about its usefulness and limited insights. I believe the purpose and scope of the manuscript are clearer without it. The analysis considers three health outcomes (SRH, IADL, ADL) stratified by age, race, and education—by sex. Previously, the analysis was also stratified by COD. Since COD is not a population characteristic as age, race, sex, and education are, it ultimately was out of place.

2. Age categories: There are now three age categories (65-74, 75-79, 85+), instead of two (Reviewer 2).

3. Educational attainment: I combined the categories of “high school” and “some college” since previous results for this older population were similar and muddled some of the patterns.

4. Table and Figures: The original Table 1 (distribution of characteristics) has been updated with the adjusted age and education variables. I switched Table 2 (health outcomes for subgroups by time to death) with the Figures (graphical representations of health outcomes) from the Appendix, since the Figures tell the story more succinctly than the Table.

Significant changes to the text include 1) the expansion of the background section to provide more context and better situate the present study; 2) the revision of the results section to better describe findings; and 3) the expansion of the discussion and conclusion sections to more clearly illustrate the paper’s contributions and implications.

I am hopeful that these revisions will satisfy the reviewers’ concerns about the original manuscript. Thank you for considering my manuscript for publication in Plos One.

--------------------

Comments from the Editor:

Both reviewers would like to see a better contextualization of the study in the existing literature, and more justification of some of your analytic decisions, for example: Why 6 years? Why these cause of death groupings? In addition, please address further the important limitation of the NHIS exclusion of the institutionalized population, and the selection effect that presents. Please explain what is meant in the Limitations section where it says that the newly-institutionalized NHIS respondents "remain in the sample" since the NHIS is cross-sectional. Also, please explain. the decision to exclude Hispanics and limit comparisons to non-Hispanic whites and Blacks.

AR: Thank you for this summary. I have clarified the decision to exclude non-white and non-Black respondents in the methods section:

“Due to small sample sizes for non-Black and non-white racial/ethnic categorizations, I limit the racially-stratified analyses to Black-white comparisons. However, the other comparisons, such as by educational attainment, include all respondents, regardless of reported race/ethnicity.”

I have also clarified the language around institutionalization:

“The NHIS does not interview individuals who at the time of interview live in a long-term care facility, though respondents who enter a care facility at some point after their NHIS interview are still linked to their death certificate at death. In other words, a non-institutionalized respondent can be interviewed by the NHIS before moving into long-term care some time later and remain in the follow-up group...”

Your other points are addressed in my response letter below.

--------------------

Reviewer #1:

Thank you for the opportunity to review this research article that considers trends and disparities in quality of life prior to death among adults age 65 and older in the U.S. The author investigates self-rated health (SRH) and disability in the last six years of life across: 1) time, 2) population subgroups (i.e., Black-White and male-female), and 3) causes of death. The author uses data from the 1987-2014 National Health Interview Survey, linked to death records through 2015, for respondents who died within six years of being interviewed. Decedents are classified as having a disability prior to death based on needing help with at least one of six ADLs and/or needing help with routine needs, such as grocery shopping or doing light chores. Causes of death are classified as: 1) accidents, 2) cancers, 3) cerebrovascular diseases, 4) chronic lower respiratory diseases, 5) a combination of heart disease and diabetes, and 6) all other causes. The manuscript is fairly well written and addresses an important topic (i.e., trends and disparities in health at end of life). Findings from such a study have the potential to inform interventions and guide program planning and address disparities. The author indicates that this study is the first known study to investigate trends in SRH in the last years of life which can make an important contribution to the literature. Some concerns noted below limit the potential impact of the paper.

1. The author should be more clear regarding data that are used. The abstract and Findings refer to data from the 1987-2014 National Health Interview Survey (NHIS) whereas the Methods (p. 5) refer to data from the 1997-2014 NHIS.

AR: Thank you for pointing out this inconsistency. I use data from 1986-2014 for SRH, and 1997-2014 for limitations (since data is not available for these variables for earlier years). I have now clarified this throughout the manuscript and most explicitly in the methods section:

“… The NHIS began consistently asking respondents about ADL and IADL limitations in 1997, but has included a question on SRH every year since the 1970’s. Since respondent matching to death records did not begin until 1986, however, and because a different weighting scheme was used in 1986, I begin the analysis of SRH trends in 1987. The analysis of ADL and IADL measures begins in 1997. I examine all outcomes until 2014, the last year for which interviewed respondents in the public-use files have been linked to death records (through 2015).”

2. The rationale for using a timeframe of up to six years-to-death is unclear. How was the six years-to-death timeframe determined?

AR: This point was also made by Reviewer #2. A longer timeframe provides more time to observe differences between groups. While a window longer than six years would have been even better, a choice of six years keeps the time trend portion of the analysis from being too outdated (since each year requires the full follow-up time to have elapsed). I have added a discussion of this balance to the Methods section:

“I assign each decedent a value for years to death by subtracting the calendar year in which a respondent was interviewed from the respondent’s year of death reported in the linked mortality file. The maximum length of follow-up before death is an important analytic decision. While some of the studies cited in this paper consider the last one or two years of life (Gill et al. 2010; Liao et al. 1999; Smith et al. 2013), others extend 3-8 years before death (Beltrán-Sánchez et al. 2016; Cutler et al. 2013; Lunney et al. 2018; Raab et al. 2018; Wilson et al. 2007), and some well beyond 10 years (Alley et al. 2010; Gerstorf et al. 2013; Stenholm et al. 2014). Lunney et al.’s (2018) findings that racial disparities in disability are ‘erased’ in the last 1-1.5 years of life suggests that a period longer than 2 years before death is needed to capture evolving patterns of disparities. The authors also find that study participants who died were more disabled at the study’s baseline three years prior to death than similarly-aged peers who did not die. This suggests that the onset of disability, an important outcome in the present study, likely begins earlier than the last few years of life.

In this paper, I consider the last six years of life. Since the annual trends portion of the analysis requires six years to have elapsed since interview and mortality data is only available through 2015, the last year for which estimates can be produced is for respondents interviewed in 2008. Although a window longer than six years would be optimal, it strikes the balance between observing outcomes and disparities for as long as possible while tracking relatively recent trends. …”

3. Given the focus on disparities among subgroups, including investigating Black-White differences, the author might consider an earlier paper by Liao et al., (1999) that also used NHIS data and can provide important context for the current paper. Using data from the 1986-1994 NHIS, linked to death records through 1995, these researchers investigated Black-White differences in disability and mortality in decedents age 50 years and older who died within 2 years of their interview. Researchers found that adjusting for educational attainment (used as a proxy for socioeconomic status) did not eliminate Black-White differences in disability and morbidity in the last years of life but did account for much of the difference. The current paper considers differences in health status with regard to race as well as education and gender but does not adjust for any potential confounding by socioeconomic status or other social determinants with regard to Black-White differences which might be important in informing interventions to address modifiable factors associated with these determinants.

AR: Thank you for this excellent suggestion. I now cite this important paper throughout the manuscript. As you suggest, the authors findings are also helpful for addressing modifiable factors. I discuss this in the discussion:

“Previous work finds that unequal access to formal education has a significant influence on end-of-life inequalities (Liao et al. 1999; Warner & Brown 2011) and adds another item to the long list of benefits to expanding educational access. Minimizing disparities in educational outcomes is a long-term approach to reducing disparities at the end of life. …”

4. A seminal paper by Lunney et al. (2003) that also is not cited by the authors investigated differences in functional decline among four types of illness trajectories: 1) sudden death, 2) cancer death, 3) death from organ failure, and 4) frailty. This research found that patterns of functional decline or disability varied substantially based on illness trajectory. Similar to the current paper, decedents who died suddenly or who died from cancer experienced the shortest periods of disability prior to death. Given that variations in the shape of disability trajectories by disease type is well established, findings related to cause of death in the current paper are not surprising or particularly compelling. Examining differences related to number of chronic conditions and their disabling effects at end of life might be more informative, especially since some evidence indicates that these disabling effects (i.e., based on number of chronic conditions) seem to be similar for Blacks and Whites who are approaching death.

AR: This comment and several of your following points prompted me to reexamine the usefulness of the manuscript’s cause-of-death analysis, ultimately choosing to eliminate it. A significant aim of the analysis is to understand end-of-life trajectories for certain subpopulations. While the other stratifying characteristics (age, sex, race, educational attainment) are about membership in certain social groups, cause of death is of course difficult to know before death. I believe the analysis is more focused without this excursion and have adjusted the manuscript and tables as necessary.

The idea to examine groups by number of conditions is compelling in light of the finding that racial disparities at the end-of-life are driven by higher order limitations. I’ve noted this in the discussion section:

“Black adults (particularly women) not only require care for longer, but require more intensive care. I find that the majority of racial disparities in one or more ADL limitations is driven by adults reporting three or more disabilities (Appendix Table 2). Future work should examine which chronic conditions are drive these racial disparities in higher order disability.”

5. Related to # 4 above, a large percentage of the sample (43% of females and 51% of males) died from accidents and cancers (shown in previous studies to be associated with the shortest periods of disability prior to death) and 41% of females and 35% of males are classified as “all other” causes of death. Only about 16% of females and 14% of males are grouped into specific non-cancer disease categories. It also is unclear why decedents with heart disease and diabetes are grouped into one category, even if people with diabetes are at high risk for heart disease. These cause of death categories do not seem particularly meaningful or informative.

AR: Please see response to comment #4.

6. Related to #4 and #5 above, the author does not indicate how overlaps in chronic conditions are handled. Given that multiple chronic conditions are common among people age 65 and older, it is likely that many decedents reported more than one chronic condition.

AR: Please see response to comment #4.

7. Related to #4, #5, and #6 above, the Liao et al., (1999) paper (referenced above under #3) that also used NHIS data to investigate disparities and trends in disability in the last years of life excluded decedents who died from accidental causes. The author might want to also consider excluding these decedents. They do not seem theoretically relevant. It also would be useful for the author to better describe how information about chronic conditions is collected in the NHIS. What are the range and types of conditions reflected in the survey?

AR: Please see response to comment #4. I do not use any data on chronic conditions. Rather, I use only data on reported difficulty with IADL or ADLs.

8. On p. 15, the authors indicates not including Alzheimer’s disease as a cause of death in the analysis based on the high likelihood of institutionalization with disease progression. However, the author indicates on p. 15 that decedents who enter a care facility after their interview are retained in the sample. Again, the rationale for the cause of death categories used in this study is questionable.

AR: Please see response to comment #4.

9. In addition, it would be useful to have more information regarding the measure of IADLs used in this study beyond just grocery shopping or doing light chores. Help with grocery shopping can be related to a lack of transportation as opposed to disability. The six ADLs are listed but less information is provided about the IADL measure.

AR: Unfortunately, the NHIS does not include additional information as to the IADL task with which a respondent requires help. Rather, the IADL measure is a simple yes/no question about whether a respondent requires any help with any IADL. I have clarified this in the Methods section:

“… The NHIS ascertained information on IADL limitations in the NHIS using a single yes/no question for whether a respondent needed help from others for ‘handling routine needs, such as everyday household chores, doing necessary business, shopping, or getting around for other purposes.’ While the above ADL variable combines six survey questions, the IADL variable reflects only this single question.”

Reviewer #2: This well-written descriptive paper provides a detailed statistical portrait of health and disability at the end of life, with careful attention to differences therein on the basis of sociodemographics (age, sex, race, education) and cause of death. The analyses are carefully done and use excellent NHIS data. Despite these strengths, the author could do more to justify their study aims and their key analytic decisions. I also encourage the author to say more about the possible influences of age vs. cohort effects, and selective survival when interpreting their results. I hope these comments are helpful to the author as they further develop this important project.

1. Why does the analysis focus on the final six years of life? Please provide a brief rationale for this decision. Other time points, such as last year of life, may align more closely with the literature on end-of-life medical expenditures, for instance. This shorter observation would help the author to better contextualize this in the literature, and link their findings to topics like expenditures. I suspect that the six-year decision was driven by sample size, but there may be other more compelling motivators.

AR: Thank you. A similar point was made by Reviewer #1. Please see my response to their comment #2 above.

2. More generally, the author could make a much more compelling case for the study goals. Why is this descriptive analysis helpful? What does it tell us, and how can these results advance research and policy/practice on end-of-life care?

AR: This point was echoed by the Editor. I have significantly rewritten parts of the introduction, discussion, and conclusion to better highlight my findings and make clearer recommendations for future research and policies.

3. PLOS is for a general readership, so more context is needed for readers who are not specialists in end-of-life topics. I would suggest a brief discussion of the strengths and weaknesses of using time-to-death measures to characterize end-of-life, and the relative strengths and weaknesses relative to measures like proxy reports and expenditure data that are other ways to characterize the end-of-life experience.

AR: Thank you for the helpful suggestion. I have rewritten part of the background section:

“From evaluating the financial wellbeing of pension systems to predicting a population’s healthcare needs, the end-of-life period is of interest across disciplines. A significant analytic decision in end-of-life studies is whether to measure age since birth, as is most typical, or to measure backward from the other end of the lifespan: death. The usefulness of a variable for remaining lifetime was first described in the 1970’s (see Sanderson & Scherbov 2013 for a history of the variable, as well as a demonstration of using the variable to study population aging). Years to death is a proxy for the complex and interacting social, behavioral, environmental, and genetic processes that determine each individual’s moment of death. An allure of the variable is that it is still under-explored, despite yielding new perspectives that are missed when using only chronological age.”

4. The section on inequalities in healthy aging, again, could do more to convey the importance and value of the research. This might be an opportune place to discuss the possibility of selective survival, such that lower SES and Blacks who survive until age 65 may show some health advantages relatives to higher SES and Whites who survive. This might be another interpretation of the seemingly counter-intuitive results whereby the least educated group appear to fare better than their more educated counterparts.

AR: Discussing selective mortality is a very helpful suggestion. I now do this in the inequalities section as you suggest:

“Another possibility through which inequalities may diminish is selective mortality. Since some populations are exposed to systematically higher mortality rates throughout their lives, these groups can be highly select by the time they reach the ages under study. By nature of their design, studies using chronological age (comparing 80 year-old white adults to their 80 year-old Black peers, for example) must ignore the influence of selective mortality. The issue is greatly lessened when considering time-to-death (comparing racial differences five years before death, for example), but nevertheless persists anytime a study sample has a minimum age below which differential mortality occurs.”

I also reference it again in the section on limitations.

“Another bias to consider when comparing differences across groups is selective mortality. Since the analysis only includes adults who survive to age 65, the estimates of between-group differences are likely smaller than they would be if everyone survived to age 65.”

5. I would suggest using a more nuanced measure of age in the analysis, such as 65-74, 75-84, 85+. The current two category measure is quite coarse and some nuanced patterns may be concealed.

AR: Thank you for the suggestion. I have updated the age categories to the ones you suggest. The new age ranges do indeed offer a more nuanced understanding and I have updated the manuscript accordingly.

6. A minor issue. On page 13, the phrase “perceived amount of time individuals spend in unfavorable health…” is misleading. Study participants were not administered perceived life span or perceived probability of survival questions. Please re-word so that the sentence more accurately characterizes the data.

AR: Good point. I have removed the word “perceived” and checked that similar phrasing is not used throughout the manuscript.

7. The analyses should raise issues of age versus cohort effects (and even touch on period effects) more directly, in the background and discussion. Even though the time period is fairly narrow, the youngest participants in 2014 and the oldest participants in 1987 represent vastly different cohorts, who have had different exposures to medications, public health interventions, assistive devices, etc. over the life course. At the very least, the limitations could say more about the inability to discern age vs cohort effects using repeated cross-sectional data.

AR: Thank you for this excellent point. In the background section where I discuss trends, I touch on changing population characteristics and environments. I now explicitly link this discussion to period vs. cohort effects:

“Because of these simultaneously evolving environments and population characteristics, it is difficult to separate period and cohort effects.”

I now also refer to the issue again in the limitations section:

“Finally, a limitation of repeated cross-sectional data is that it is not possible to distinguish between period and cohort effects. Measuring the relative importance of cohort composition vs. period changes in the treatment of illness and disability would help target relevant interventions.”

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

2 Feb 2022

3 Mar 2022

Response to Reviewers

Dear Professor Idler and Reviewers,

I would like to thank you again for the time you have taken to provide constructive feedback. As before, I below respond to each point with AR (for “Author’s Response”).

Best,

The Author

Reviewer #1

Thank you for the opportunity to review this revised manuscript that considers trends and disparities in end-of-life health in the U.S. The author made a good effort to address both reviewers’ (and the editor’s) comments/concerns and has strengthened the paper. Below, I have outlined some additional points to consider. Some additional editing and wordsmithing are needed and would further improve the quality of the paper.

1. The abstract could use some editing. For example, the sentence: “Time spent in fair/poor health over years 1987-2008 declined by two months, while time lived with at least one activity limitation generally remained stable from 1997-2008” is a little clunky. Time spent in fair/poor health over years 1987-2008 declined by two months compared to what? Declined by two months each year? The sentences that follow also could use some editing. The sentence: “Compared to men, women reported an IADL for 1 year longer and an ADL for 8 extra months, yet both sexes reported similar lengths of unfavorable health” also needs editing. The numbers 1 and 8 should be spelled out. Reported an IADL or ADL what? Limitation? what does the author mean by “similar lengths of unfavorable health? The way that the sentence is currently written, it is not clear what is meant here. In the next sentence, I suggest saying similar health “compared with” younger decedents instead of “to” younger decedents. Given the use of repeated cross-sectional data, I suggest using language such as “findings indicate” rather than stating the findings as if they are fact in the Discussion section of the Abstract.

AR: Thank you for these suggestions. I have significantly re-written the Results and Discussion sections of the abstract.

2. On page 4, the author indicates the importance of examining ADL limitations (e.g., 40% of people over age 65 have at least one limitation and nearly 90% with 3+ limitations require caregiving help) but do not say why it might be important to also consider IADL limitations. They do make the case on p. 10 and may want to say something similar here on p. 4 as well. The sentence: “ADL limitations are predictive of requiring physical assistance, with roughly 40% of community-dwelling adults age 65+ with one limitation and nearly 90% with 3+ receiving caregiving help” needs editing to read more clearly. Also, physical assistance is caregiving help.

AR: In addition to clarifying the suggested sentence, I have added a sentence to expand on the utility of IADL limitations:

“While IADL limitations are less disabling than ADL limitations, an IADL limitation indicates that an individual requires some level of support in order to live independently.”

3. On p. 5, what is meant by the sentence: “Because of these simultaneously evolving environments and population characteristics, it is difficult to separate period and cohort effects.” I am not following this line of thought. I agree that the methods used in this paper (and acknowledged under Limitations) limit the ability to disentangle these effects.

AR: I have re-written the sentence to read:

“Because cohorts are evolving at the same time as the contexts in which they live, it is difficult to separate period and cohort effects.”

4. On p. 7, instead of using wording such as, “an older study similarly constructed to this one,” consider, for example, referring to “a similar study” or “another study with a similar design,” etc. The wording here is clunky.

AR: I have made the change to “a similar study”.

5. In describing the sample on p. 9 in the Methods section, information should be included regarding any exclusion criteria (e.g., living in a long-term care facility at baseline). Information should also be included here regarding the possibility that those not in long-term care at baseline could be included at follow up.

AR: I have added the following sentence:

“While residents of long-term care institutions are not included in the baseline sample, the sample may include individuals who were interviewed at home and then moved to an institution during the follow-up period.”

6. The literature reported in the first paragraph under the section on "Years-to death” on p. 9 in the Methods section feels out of place here. This literature should be woven into the Background section that begins on p. 3 and not here. A Methods section typically does not include a literature review. The rationale for the six years-to death timeframe can be made without including all of these citations in the Methods section.

AR: As you suggest, I have moved the section with the citations to the Background section, while maintaining a shortened rationale for the time frame in the Methods section.

7. On p. 12, I suggest more clearly outlining how data were pooled to conduct the varying analyses and describing the different analytic approaches in turn rather than stating: “To compare estimates across population subgroups, I pool data across years 1997-2014 (1987-2014 for SRH).”

AR: I have rewritten this sentence and avoided using the term “pooling”, which made the approach sound unnecessarily complicated. The sentence now reads:

“To compare outcomes across population subgroups, I estimate the mean of an outcome prevalence across the study period (years 1997-2014 for SRH; years 1997-2014 for IADL and ADL).”

8. Avoid using language like, “women reported either kinds of limitations for much longer” on p. 14. Again, the paper could benefit from some wordsmithing throughout to improve readability and flow and to sound more academic.

AR: I have adjusted the wording and gone through the manuscript to find a balance between accessibility and sounding academic.

9. On p. 21, avoid using the term “elder.” The term, “older adults” is preferred by aging researchers and others.

AR: I have the term changed “elder care” to “care for older adults.”

10. When referring to Blacks and Whites, I suggest capitalizing both terms to be consistent.

AR: Thank you for this suggestion. I gave careful thought to my decision to capitalize Black, but not white, in my past submissions of the manuscript. The stylistic debate today is of course tied to discussions around identity and racism in the US today.

Around the time of my first submission in 2021, the Associated Press announced that it would capitalize Black, and not white.

This thought piece in The Atlantic also gives a good overview of the current debate.

On reconsideration of the current debate (and in particular this piece by Dr. Eve Ewing), I have chosen to capitalize White.

The initial aim was of the typographical inconsistency was to highlight the upper case B. A prominent argument against the lowercase w, however, is that its use doesn’t recognize Whiteness as a sociological construct and allows White people to sit out on the debates and work around racism.

Therefore, I have made the change throughout.

11. Throughout the body of the text, “vs.” should be written out as “versus” unless included in material with parentheses.

AR: Thank you. I have made two necessary changes.

12. On p. 11, instead of “non-Black and non-white racial/ethnic categorizations,” I might indicate that sample sizes for other racial and ethnic groups were too small to conduct any additional racial/ethnic comparisons. The language used here is a little clunky.

AR: Good point. The sentence now reads:

“Due to the small sample sizes for other racial/ethnic categorizations, …”

13. On pp. 8-11, text indicates that different sample sizes were used for different analyses (e.g., SRH vs. ADLs and IADLs as well as racial comparisons vs. other types of comparisons (e.g., educational attainment). It is not clear what the different sample sizes were for the various analyses.

AR: Thank you for pointing this out. The information was previously buried in the notes for Table 1. I now mention this in a footnote in the section describing the sample size:

“The SRH analysis sample consists of 77,295 individuals across all years 1987-2014. Sample sizes for the IADL and ADL analyses, which span years 1997-2014, are 40,354 and 40,359, respectively.”

14. On p. 23, The author indicates that “measuring the relative importance of cohort composition vs. period changes in the treatment of illness,” etc. would help target relevant interventions.” A brief example of how this type of analysis could inform relevant interventions might be useful here. As noted by Reviewer 2, the methods used and inability to disentangle age versus cohort versus period effects is a limitation that should be acknowledged. The author acknowledges this limitation but does not provide a concrete explanation of how/why the analysis is limited because of this constraint.

AR: Thank you for the suggestion. I have added the second sentence:

“Measuring the relative importance of cohort composition versus period changes in the treatment of illness and disability would help target relevant interventions. For example, knowing that older male cohorts are less likely than their predecessors to be heavy smokers (Preston & Wang 2006), but more likely to face the health problems associated with obesity (Wang et al 2011), could inform the decision to divert funds away from tobacco-related interventions and toward new programs targeting obesity.”

15. In the conclusion on p. 23, I suggest saying something like, “findings indicate”, etc., rather than stating results as fact.

AR: I now use weaker language throughout the discussion.

Reviewer #2

The authors have done a very thorough and responsive revision. I have just a few remaining suggestions for improvement.

1. The manuscript would benefit from a very careful copy-edit both for clarity and style. For instance, the paper opens on an awkward note. The first sentence of a manuscript should be more direct, such as “Death—and the months, days, or moments preceding it—are an important and distinct stage of the life course (Cohen-Mansfield et al. 2018)” (deleting initial clause).

AR: Thank you for the suggestion. I have adjusted the first sentence and have made clarifying changes throughout the manuscript.

2. A recent paper by Carr and Luth (2019) Annual Review of Sociology makes the case that ‘end of life’ is a distinctive life course stage. This article could be helpful for your framing of the analysis.

AR: I now cite this helpful paper throughout the manuscript.

3. The front end of the paper could do a bit more to motivate the selection of the multiple health measures. These issues are addressed somewhat in the Discussion, but it would be helpful to foreground the fact that SRH is subjective and assessed in relation to one’s peers (who may be dying or in poor health), whereas the IADL and ADL measures are more ‘objective’ and reflect behavioral capacities.

AR: Thank you for this suggestion. The Background section now reads:

“SRH is a subjective and self-reported indicator of health. While disability is also self-reported, it serves as a more objective measure of requiring assistance.”

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

12 Apr 2022

How does it all end?

Trends and disparities in health at the end of life

PONE-D-21-16309R2

Dear Dr. Vierboom,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Ellen L. Idler

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

One of the reviewers for this round is still requesting additional line-editing before final submission, for precision and more use of active voice. Please see the reviewer comments below.

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: The authors have done a highly responsive revision, and the main methodological and conceptual concerns have been adequately addressed. I have lingering concerns about the writing, which could be more direct and precise in places. I would encourage the use of active v. passive voice where possible, and avoiding the use of vague phrases that add little to the text. For instance, rather than referring to "various populations" just specify precisely what the populations are. No need to use phrases like "in this paper" (as it is clear that the proposed analyses are being done in your paper).

I encourage a very careful line-edit to enhance the clarity and conciseness of the work. Congratulations on a successful revision.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

30 Jun 2022

PONE-D-21-16309R2

How does it all end? Trends and disparities in health at the end of life

Dear Dr. Vierboom:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Ellen L. Idler

Academic Editor

PLOS ONE