Sage Journals: Discover world-class research

Abstract

We examined regression models predicting health services standardized costs (HSSC) during the years preceding death using varied temporal parameters related to the dependent and independent variables. The regression models sought to elucidate how costs before the final year of life, temporal factors, and demographics are associated with costs in the final year. Anonymized data were derived from the records of Israel’s largest health maintenance organization for 71,855 people aged 65+ in 2006, who died between 2008 and 2011. In the regression models, the Independent Variables of age, sex, and comorbidity (as measured by the Charlson Comorbidity Index) were significant predictors of the dependent variable of HSSC during the final year of life. However, the strongest predictor (independent variable) of the dependent variable, HSSC in the final year of life was the independent variable, HSSC in the years preceding the final year of life. Prediction was more accurate when the predicting period was closer to the predicted period. Accuracy declined as the predicted period approached death. The results provide insights into methodological considerations in the process of prediction of end-of-life expenditures, which may assist in setting methodological standards that may facilitate arriving at consistent findings in this field. While end-of-life is associated with aberrant increases in costs, that is, increases that deviate from prior predictions, significant predictions can still be made.

Keywords

health services standardized costs end-of-life death aging comparison of multiple hierarchical regression models

Introduction

Health Services Standardized Costs (HSSC) rise at the end-of-life.^1
-3 The final year of life accounts for about 25% of Medicare expenditures for older persons.^4
-6 In Israel, persons older than 65 years, who constitute about 11% of the population, were estimated to account for 31% of Israel’s annual public healthcare expenditure in 2014.⁷ Given the pressure to manage such costs,^8
-10 examining the factors that explain end-of-life healthcare expenditures is crucial. Therefore, we examine the explanatory power of HSSC in the years preceding end-of-life in the prediction of HSSC in the final year of life via alternative regression models.

Israeli citizens are entitled to universal acute and rehabilitation care through one of four health maintenance organizations (HMOs). Based on acute and rehabilitative care cost data from Israel’s largest HMO, Clalit Health Services (Clalit), we tested HSSC’s explanatory power in conjunction with other predictor variables of end-of-life health expenditures: age, sex, socio-economic status, medical status, and length of residence in Israel. We investigated whether temporal factors, like length of predicting period (the period to which the study’s independent variables pertain), or predicted period (the period to which the dependent variable pertains), or the time proximity of the predicted period from death, altered explanatory strength.

While factors, including age, time-to-death, sex, healthcare system “navigation skills,” and socio-economic status have been found to significantly impact HSSC, contradictory findings exist. Some studies have found that HSSC increases with aging,^11,12 while others have indicated that length of time to death, and not age itself, is associated with increased expenditures.^13
-15 Still others have found that costs decrease in the final year of older patients’ lives.⁴ Some evidence suggests that the relationship between HSSC and age may not be linear. One study found that expenditures increased with age for non-decedents (until about age 90, after which expenditures declined), but expenditures for decedents aged 65+ followed an inverse U-shaped curve, rising until around age 74, and then declining.⁴

The literature regarding the association between sex and HSSC during the final year of life has been inconsistent. While some studies reported that women’s expenditures exceeded those of men in their final year,^16,17 this may have been applicable only to youngest older adult cohorts (ages 68-79), with the reverse applicable to oldest cohorts.¹⁸

An Israeli study found that different socio-economic groups presented distinct HSSC patterns. Lower socio-economic individuals made more use of emergency departments and hospitals while higher socio-economic individuals made more use of ambulatory services and prescription drugs.¹⁹ Another study reported that established Israelis had more hospitalizations and emergency room visits compared to individuals who immigrated even more than 10 years before the study.²⁰

Much research investigating end-of-life HSSC has focused solely on the final year,^21
-24 or half year or even the last 3 months.^25,26 However, some studies have investigated health expenditures over longer periods before death, finding that discrepancies in health expenditures may begin before the final year.^12,18,27 One study reported different expenditure patterns during the final 3 years for different ages, sexes, and other background characteristics.¹⁸ Another reported that HSSC and other expenditures differed between decedents and non-decedents even 6 years before death.²⁷

In their recent review of end-of-life expenditures, Kocot et al²⁸ concluded that the results of prior studies have been largely inconsistent. This study seeks to expand the understanding of HSSC during the last year of life by: (1) investigating these costs in relation to expenses incurred in the years preceding the year of death; (2) examining the influence of temporal factors on cost prediction, such as the time interval between the prediction period and the reference period for independent variables, as well as the proximity of these periods to the time of death; and (3) clarifying the effects of health and demographic variables for which previous research (summarized above) has found inconsistent results concerning their impact on health service end-of-life costs. We carried out the study using a large and robust dataset. As to the first two aims, our study provides methodological clarifications in the pursuit of understanding determinants of end-of-life expenditures. With respect to the last aim, we delve further into areas where past research yielded inconsistencies.

Based on the literature we reviewed, we hypothesize that temporal proximity to death of health cost data significantly explains healthcare costs during the final year of life. The demographic, social, and health variables included in our model are conceptualized through Andersen’s Behavioral Model of Health Services Use.²⁹ Specifically, predisposing factors are represented by demographic and social variables (eg, socio-economic status and length of residence in the country), while the need factor is captured by patients’ comorbidities, with the healthcare organization remaining constant, as all participants were enrolled in the same HMO. While these variables are incorporated into the regression models to account for service costs, this study also includes prior-year costs and the timing of background and cost variables in order to explore their impact on predicting health service costs.

Methods

Design, Setting, Materials, and Participants

This study examines the utility of Health Services Standardized Costs (HSSC) in years before death in predicting HSSC in the final year after controlling for other predictor variables: age, sex, socio-economic status, medical status, and length of residence in Israel, via comparisons of multiple hierarchical regression models in which we varied predictors and temporal factors.

HSSC refers to acute and rehabilitative care costs in relation to a designated standard of such costs. The standard we used was health care costs of individuals in the study population who did not die in immediately following years. HSSC was calculated as the total acute and rehabilitative care cost per individual per quarter divided by the average of the median acute and rehabilitative care costs per individual per quarter for 2006 among individuals who remained alive at mid-2012.³⁰ This may be summarized as:

H S S C_{i, y, q} = \frac{C o s t_{i, y, q}}{\frac{1}{4} \sum_{q = 1 - 4} M e d i a n C o s t_{2006 . q}}

where i is the individual, y is the year of study, and q = 1, . . ., 4 are the quarters.

The dataset was obtained from Israel’s largest HMO, Clalit Health Services. The study was approved by the Helsinki Ethics Committee of Carmel Hospital of Haifa, Israel, an affiliate of Clalit Health Services, approval 0015-12-COM on August 27, 2012. The study was exempted from the requirement of obtaining informed consent since the dataset was anonymized before being released to the authors.

The dataset we received from Clalit was limited to the variables examined in this paper, that is, information about HSSC, demographics and comorbidity by date for all individuals who were included in the dataset. Selection criteria were decedents who were over age 65 in 2006 and who died between 2008 and 2011 (N = 71,855). Data for those who remained alive (N = 48,482) in those years were included in the dataset, but not used in the regression models presented in this paper. HSSC, comorbidity, and demographic data were collected for years 2006 to 2012 in order to obtain at least 2 years of data before date of death.

Measurements

The measurements chosen focused on variables in the dataset thought to impact Health Services Standardized Costs.

Demographics

Data included age in 2006, age at death, sex (female coded as 1 and male as 2), length of residence in Israel (categorized as: born in Israel, immigrated before 1948, 1948 to 1955, 1956 to 1989, and after 1990), and whether a person was exempt from health insurance copayments, Clalit’s measure of low socio-economic status, with low socio-economic status coded as 2 and other coded as 1. The correlation between age in 2006 and age at death was .997 (P < .001), and the variable, age, was calculated in relation to age at death.

Medical Data

Medical status was captured from Clalit’s chronic disease registry, using the Charlson Comorbidity Index,³¹ a measure including 19 diseases, with each allocated a weighted score based on severity, for example, AIDS and cancer have higher scores. The Charlson Comorbidity Index does not include age as a factor.

Health Services Standardized Costs

Expense items included in HSSC were the cost of acute and rehabilitative care, hospital care, primary and specialist physicians, pharmaceuticals, imaging, and laboratory tests. Because of cases of extreme costs, means were generally higher than medians. The costs of specific services were based on Clalit’s determinations. The quarterly median cost is the proprietary information of Clalit. Therefore, HSSC is presented as a standardized cost. We calculated yearly, half yearly and quarterly HSSC.

Statistical Analytic Strategy

Hierarchical multiple regressions were run to investigate the potency of background and HSSC variables in accounting for HSSC in the final year of life (dependent variable), and to study whether using HSSC in previous years as an independent variable improved the explanatory power of regressions with HSSC in the last year of life as a dependent variable. The independent variables of age, sex, length of residence in Israel, socio-economic status, comorbidity per the Charlson Comorbidity Index, and HSSC in previous years were added sequentially to account for collinearity and investigate their individual impact. To explore HSSC’s explanatory power regarding subsequent HSSC, we framed four questions, which were answered through comparison of the respective regression models. The questions examined four types of temporal characteristics of independent variables and/or the dependent variable: (1) The relationship between the explanatory power of the regression and the time interval between predictor (independent variable) and the dependent variable period: Is HSSC in the final year of life better predicted by HSSC in the prior year or by HSSC in year 3 prior to date of death?; (2) Length of predicting period, that is, the independent variable of HSSC in several years prior to the final year of life vs. the independent variable of HSSC in the year prior to final year of life: Is HSSC in the final year of life better explained by HSSC in the prior year or by HSSC in the period including all three years prior to the final year of life?; (3) Time proximity of predicted period to death: Is the explanatory power to predict HSSC in the final year of life by HSSC in the prior year thereto (ie, year 2 before the date of death) better or worse than the prediction of HSSC in the third year prior to the date of death by HSSC in the prior year (ie, year 4 before the date of death)?; (4) Length of the period pertaining to the dependent variable, for example, full year versus half year: Is the explanatory power of predicting HSSC in the first six months of the final year of life (the dependent variable) by HSSC in the prior year (independent variable) better or worse than the explanatory power of predicting HSSC in the full final year (as the dependent variable), when using the same independent variable (ie, HSSC in the first year prior to final year of life (year 2 before the date of death)? We thus explored different regression models, attempting to infer general principles regarding the parameters affecting prior HSSC’s strength in explaining later HSSC.

For the regression where the period of the independent variable was 2 years before the date of death, we studied only individuals who died from 2008 onward in order to include HSSC and Charlson Comorbidity Index scores of 2 prior years as independent variables. Likewise, we studied individuals who died from 2010 onward for the regression where we used HSSC 4 years before the date of death as an independent variable.

Due to the size of the sample, we regarded only P < .001 as significant. Regression models examine the ability of the independent variable to predict the dependent variable, yet when such prediction occurs it does not imply causation, since other mechanisms are often involved in the association.

Results

Demographic and Background Characteristics

Characteristics of participants who died 2008 to 2012 are described in Table 1. Inter-correlations among the main variables are presented in a Supplemental File.

Table 1.

Sample Characteristics of People Who Died Between 2008 and 2012 (N = 71,855).

Variable	Mean (median); %	Standard deviation
Age in 2006	78.97 (80.00)	6.34
Age at death	82.46 (83.00)	6.39
Health services standardized costs 2006 1st quarter^a	10.50 (0.79)	33.06
Health services standardized costs 2006 1st half^a	26.77 (2.36)	74.75
Health services standardized costs 2006 entire year^a	60.89 (7.55)	141.60
Sex—% female	53.9%
Low socio-economic status	39.9%
Charlson Comorbidity Index	2.76 (2.00)	2.35
Charlson Comorbidity Index groups
Low (1-2) %	35.2%
Medium (3-5)	52.2%
High (6+) %	12.5%
Age groups (of 2006)
65-74	26.1%
75-84	51.8%
85+	22.1%
Immigration group/length of residence in Israel until 2006
Israeli born	9.0%
Before 1948/58 or more years	12.6%
1948-1955/51-58 years	36.8%
1955-1990/16-51 years	26.4%
After 1990/16 or less years	15.2%

Note. ^aHealth services standardized costs was calculated as total costs per individual per quarter, divided by the average of the median 2006 quarterly costs for those who were alive by mid-2012.

The Potency of ‘HSSC in Years Before the Final Year of Life’ as an Independent Variable in a Regression Predicting HSSC in the Final Year of Life (the Dependent Variable)

A hierarchical regression was run to determine if the addition of the independent variable, HSSC one year before the period of the dependent variable would add a statistically significant explanation of the dependent variable, HSSC in the final year of life, over and above the other independent variables. The full regression model, which included the independent variables of age, sex, socio-economic status, length of residence in Israel, comorbidity and HSSC in Year 1 (0-12 months) before the final year of life (Table 2, Model 2), was statistically significant, R² = .161, F(9, 71845) = 1530.1, P < .001; adjusted R² = .161. The addition of the independent variable, HSSC in the year before the final year of life to the regression without it (Table 2, Model 1) led to a statistically significant increase in R² from .047 to .161. P < .001. The only independent variables that were not significant predictors of the dependent variable, HSSC in the final year of life were length of residence in Israel and socio-economic status. HSSC in the first year prior to final year of life (year 2 before the date of death) had a larger β than did all other independent variables.

Table 2.

Utility of Health Services Standardized Costs (HSSC) in Years Prior to Final Year of Life (IVs) as a Predictor of HSSC in Later Years (DV).

Model typeVariable	Model 1: without HSSC predictor	Model 2: with HSSC predictor	Model 3: earlier predictor period	Model 4: earlier periods of both predictor and predicted	Model 5: earlier period of predicted period
Predicted period	Final year of life	Final year of life	Final year of life	Year 3 prior to death	First half of final year of life
Predicting period	Year prior to final year of life	Year prior to final year of life	Year 3 prior to predicted period	Year 4 prior to death	Year prior to final year of life
Dependent variable	HSSC in the final year of life	HSSC in the final year of life	HSSC in the final year of life	HSSC in Year 3 prior to death	HSSC in the first half of the final year of life
N	71,855	71,855	35,644	35,644	71,855
Independent variable	β	β	β	β	β
Age	−.177*	−.138*	−.174*	−.068*	−.077*
Sex (ref = female)	.028*	.021*	.029*	.009	.013*
Socio-economic status (ref = high SES)	−.006	−.008	.007	−.003	−.004
Charlson Comorbidity Index	.102*	.046*	.058*	.099*	.047*
Immigration year/length of residence in Israel (ref = Israeli born)
Before 1948/58 or more years in Israel	−.015	−.008	−.007	−.012	−.003
1948-1955/51-58 years in Israel	−.022	.009	.009	−.011	−.005
1955-1990/16-51 years in Israel	−.004	.006	.007	−.013	−.004
After 1990/16 or less years in Israel	−.028*	−.016	−.012	.021	−.015
Yearly HSSC In predicting period	—	.344*	.177*	.392*	.395
Adjusted R²	.047	.161	.079	.191	.180

Note. N in the regressions is different due to the different number of years included in the regression.

IV = independent variable; DV = dependent variable.

P < .001.

The Relationship Between the Explanatory Power of the Regression and the Time Interval Between Predictors (Independent Variables) and the Dependent Variable Period

We compared our original regression model, Table 2, Model 2, to a model in which the independent variables were from Year 3 (24-36 months before final year of life), Table 2, Model 3. The latter model showed that these independent variables significantly contributed to the regression equation explaining the dependent variable (ie, as compared to Table 2, model 1), HSSC in the final year of life, R² = .079, F (9, 35634) = 339.08, P < .001.

Age was negatively associated with HSSC in the final year of life (β = −.174), that is, older decedents showed lower HSSC at end-of-life relative to younger decedents. Sex and Comorbidity were significantly associated with HSSC, but β-values were low, indicating less explanatory power of HSSC in the last year of life as compared to age. Males incurred higher end-of-life HSSC than females (β = .029), and the higher a participant’s comorbidity score, the higher the end-of-life HSSC (β = .058). Length of residence in Israel and socio-economic status in Year 3 prior to the last year of life were not significantly associated with HSSC for the last year of life.

In the original regression model (Table 2, Model 2) and altered regression model (Table 2, Model 3), the independent variable of HSSC in years prior to the final year of life was a significant predictor of the dependent variable, HSSC in the final year of life. However, HSSC’s explanatory power was higher in the original (higher β-value) when the interval between the predicting and predicted period was shorter. Whereas the independent variable, HSSC in the years prior to the last year of life was clearly the most potent predicting variable in the original model using data from the year prior to the final year of life to predict HSSC in the final year of life, in the model using Year 3 data, age was almost as strong a predictor (β = .177 for HSSC, β = −.174 for age). The explanatory power of the variables of socio-economic status, sex, length of residence in Israel, and comorbidity were similar in the two regression models.

Length of Predicting Period

We compared the original regression model (Table 2, Model 2) to a model that included HSSC for the combined three years before the final year of life as an independent variable. Similar R² values in the original and adjusted regression models (not shown in the tables) indicated that using a longer predicting period offered no explanatory advantage over a shorter period (R² = .163, F(9, 35634) = 770.5, P < .001).

Time Proximity of Predicted Period to Death

To determine whether time proximity from death would affect the explanatory power of the independent variable, we compared our original regression model (Table 2, Model 2) to a model in which the dependent variable was HSSC in Year 3 before the date of death and the forecasting (independent variables) period was Year 4 before the date of death (Table 2, Model 4), that is, we predicted HSSC in Year 3 before the date of death based on independent variables measured in the previous year (Year 4 before the date of death). Despite the identical time interval between predicting and predicted period in these models, that is, 1 year in both, the estimation of HSSC in Year 3 before the date of death (Table 2, Model 4) was more robust than that of HSSC in the final year of life, as evidenced by higher R² for the model predicting HSSC in the third year prior to death (R² = .190) compared to the original regression model predicting HSSC in the final year of life (R² = .161). Thus, the explanatory power of the regression, was higher when it was more temporally distant from the date of death.

Age was a better predictor of HSSC when the predicted period was closer to date of death. (β = −.138 for original model and −.068 for adjusted model). Sex was a significant predictor only in the original model (β = .021) and not the temporally adjusted regression model (β = .009). In contrast, comorbidity predicted the dependent variable better in the third year before death (β = .099) than HSSC in the final year of life (β = .046).

Length of the Period Pertaining to the Dependent Variable

We compared our original regression model (Table 2, model 2) to an adjusted model in which the outcome variable (dependent variable) was HSSC in the first half of the final year of life, that is, months minus 12 to minus 6 from death. As shown in Table 2, model 5, the R² of .180 for the first half of the year was higher than that for the whole year regression (R² = .161). The full year predicted period included data from the second half of the year, which was closer to death, where the largest variability occurs. To clarify whether this finding was specific to the final year of life, we repeated the same analyses, with the dependent variable being HSSC in the first half or the full year during the third year before date of death, with the independent variables pertaining to the fourth year before death. In this timeframe, pre-death variability was less likely to be manifested (Table 3).

Table 3.

Predicting Health Services Standardized Costs in the First Half and Full Third Year Prior to Death (Based on Data from Year 4 Prior to Death) Compared to the First Half and Full Last Year of life (Based on Data from Year 2 Prior to Death).

Predicted periodVariable	Year 3 prior to death (N = 35,644)		Last year of life (N = 71,855)
	Full year	First half year	Full year	First half year
	β	β	β	β
Age	−.068*	−.048*	−.138*	−.077*
Charlson Comorbidity Index 4/2 years, prior to death	.099*	.079*	.046*	.047*
Yearly HSSC 4/2 years prior to death	.392*	.356*	.344*	.395*
R ²	.191	.151	.161	.180

P < .001.

Table 3 displays the R² and β values of the regressions with the dependent variables being HSSC in the first half of the third year before the final year of life and HSSC in the full third year before the final year of life. These are compared to the dependent variables of HSSC in the first half year and full year of the final year of life. The differences in length of predicted period revealed differences between the third year before death and the final year of life. While increasing the length of the predicted period decreased the explained variance (R²) in the final year of life, it increased the explained variance in the third year before death. The same pattern was seen with the βs for yearly HSSC, but not for age.

Discussion

A recent review²⁸ of end-of-life expenditures on health care of older persons concluded that results of prior studies concerning such expenditures are often inconsistent and that methodological issues pose a barrier to research comparability. The research presented in our paper adds another datapoint to the inconsistent available results (eg, in terms of the contribution of sex, age, socio-economic status, number of medical conditions). This study also contributes to the understanding of methodological issues related to the temporal relationships among the independent variables and the dependent variable and the time of death.

In terms of adding another finding to the mixed findings of past literature, our study found that end-of-life health care expenditures decreased with age (within a population of older persons). Our findings indicated that it was higher for males than for females (though the effect of sex was small); care expenditures were not significantly related to socio-economic status – within a population covered by an HMO in a country (Israel) with universal health coverage; and expenditures increased with higher comorbidity. Since inconsistencies were found among earlier studies for each of these, our findings are supported by some prior studies but not by others.²⁸ However, our finding concerning age seems to be supported by the majority of prior studies.

Prior studies tended to agree that end-of-life healthcare expenditures tended to increase with increased proximity to time of death,²⁸ yet we did not find prior studies that examined the impact of prior health expenditures on expenditures in the final year of life, or on the impact of temporal factors on end-of-life health expenditures. Our regression models demonstrated the utility of using HSSC in the years preceding death to estimate HSSC in the final year of life, and the impact of various temporal parameters on the ability to predict HSSC:

(1) HSSC in years prior to death served as a significant predictor of HSSC in the final year of life. The explanatory value of HSSC in prior years was greater than other demographic and medical variables.

(2) Closer time proximity between the period pertaining to independent variables and the dependent variable increased the proportion of variance explained by the independent variables.

(3) Greater length of predicting period (of the independent variables) was not associated with explanatory value.

(4) Greater time interval between predicted HSSC-period and date of death increased the proportion of variance explained, but with varied effects on different predictors, for example, HSSC in one of the final years of life was best predicted by HSSC in prior years when the predicted period was farther from date of death.

(5) The effect of increasing the length of the predicted period depended on the time interval between the predicted period and the date of death. When the predicted period was further from the date of death, using a longer predicted period increased the explanatory power of the regression model, whereas using a period closer to death, yielded reduced explanatory power.

The results are based on specific comparisons, for example, the difference in the explanatory power between predictors measured either 1 versus 3 years before the final year of life.

The study’s value exceeds the statistical significance of its regression models. We found that the explanatory power of changing temporal factors was often more potent than the contributions of the background variables. Had we chosen other comparisons, for example, testing the difference in explanatory power between predictors measured 2 versus 3 years before the final year of life, such comparisons would have affected the specific effect size, while it was unlikely that they would have affected the principal findings. Thus, this study illustrates general trends rather than unique occurrences. Due to sharp increases in medical expenses during the final year of life, there is greater variability in final year HSSC compared to HSSC in preceding years. Closer time proximity of HSSC data to date of death reduces predictive power, perhaps explaining why the independent variables better explained HSSC when the predicted period was the third year before death rather than the final year of life.

That comorbidity significantly predicted HSSC confirms previous findings that higher comorbidity is associated with higher hospital admissions and healthcare utilization.^32,33 That socio-economic status was not a significant predictor of HSSC may be related to Israel’s provision of healthcare access to all citizens, in contrast to some other countries.^34
-36 Our results suggest that HSSC’s increase at end-of-life starts earlier² than is conventionally thought. While HSSC data closer in time to date of death provided reduced predictive power, and accounted for modest percentages in HSSC variance, using such data in hierarchical regression models represents progress in utilizing these data in comparison to previously reported models.

The main limitation is the use of an ex-post perspective, based on a sample of deceased persons, where time proximity from death is known. Another limitation is that the data pertain to a period in the past, whereas utilization of new therapies and palliative care has increased. Since long-term residential care is not covered by Israeli HMOs,³⁷ its cost was not included in calculating HSSC. A further limitation is that we had access only to the data available in the Clalit dataset. Different disease processes, like organ failure and terminal illnesses may produce distinct HSSC trajectories at end-of-life,²³ but such data were not available to us. More recent measures of comorbidity or disease burden/risk, may be of greater predictive use than the Charlson Comorbidity Index.^38,39 Future research should examine indicators of specific disease categories, of health status change, or indicators of quality of life or quality of care as predictors. Long-term care costs, like those for nursing homes, assisted living, or in-home services were not included in this study, and should be examined in further research as trends could be different for such services. Future research should also focus on the prediction of date of death, and pre-death HSSC in a prospective manner. Emerging methodological approaches^26,40 should also be tested to examine the advisability of expensive or painful procedures under various trajectories.

Conclusion

The study results are important for suggesting an examination of prior health expenditures in models predicting end-of-life expenditures, a practice not previously examined in the literature. In addition, the results provide insights into methodological considerations in the process of prediction of end-of-life expenditures, which may assist in setting methodological standards that may facilitate arriving at consistent findings in this field. Finally, the results suggest that despite a state of seemingly increasing entropy in health expenditures, significant predictors do emerge as older persons draw closer to the date of death.

Supplemental Material

sj-docx-1-inq-10.1177_00469580251326315 – Supplemental material for End-of-Life Health Costs Were Predicted Primarily by Prior Health Costs, and Secondarily by Temporal, Health and Demographic Factors

Supplemental material, sj-docx-1-inq-10.1177_00469580251326315 for End-of-Life Health Costs Were Predicted Primarily by Prior Health Costs, and Secondarily by Temporal, Health and Demographic Factors by Jiska Cohen-Mansfield, Michal Skornick-Bouchbinder and Moshe Hoshen in INQUIRY: The Journal of Health Care Organization, Provision, and Financing

Footnotes

Acknowledgements

The authors thank Clalit Health Services, Israel for enabling them to use its database for this study.

Author Contributions

JC-M conceptualized the study. JC-M and MSB wrote the manuscript. JC-M, MSB, and MH were involved in the analysis and interpretation of data and editing the manuscript.

Data Availability

The datasets presented in this article are not readily available because the authors are not the owners of the data. Requests to access the datasets should be directed to the corresponding author, Jiska Cohen-Mansfield, jiska@tauex.tau.ac.il, who will submit any data requests received to the owner of the data, Clalit Health Services.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Minerva Foundation, grant number 3158329500.

Ethical Approval and Informed Consent

This study was approved by the Helsinki Ethics Committee of Carmel Hospital of Haifa, Israel, an affiliate of Clalit Health Services, approval 0015-12-COM, August 27, 2012. The study was exempted from the requirement of informed consent because the dataset was anonymized before it was released to the authors.

ORCID iD

Jiska Cohen-Mansfield

Supplemental Material

Supplemental material for this article is available online.

References

Felder

Meier

Schmitt

Health care expenditure in the last months of life. J Health Econ. 2000;19(5):679-695. doi:10.1016/S0167-6296(00)00039-4

Seshamani

Gray

AM.

A longitudinal study of the effects of age and time to death on hospital costs. J Health Econ. 2004;23(2):217-235. doi:10.1016/j.jhealeco.2003.08.004

Chastek

Harley

Kallich

Newcomer

Paoli

Teitelbaum

AH.

Health care costs for patients with cancer at the end of life. J Oncol Pract. 2012;8(6S):75s-80s. doi:10.1200/JOP.2011.000469

Lubitz

Riley

GF.

Trends in Medicare payments in the last year of life. N Eng J Med. 1993;328(15):1092-1096. doi:10.1056/NEJM199304153281506

Riley

Lubitz

JD.

Long-term trends in Medicare payments in the last year of life. Health Serv Res. 2010;45(2):565-576. doi:10.1111/j.1475-6773.2010.01082.x

Duncan

Ahmed

Dove

Maxwell

TL.

Medicare cost at end of life. Am J Hosp Palliat Med. 2019;36(8):705-710. doi:10.1177/1049909119836204

Dwolatzky

Brodsky

Azaiza

Clarfield

Jacobs

Litwin

Coming of age: health-care challenges of an ageing population in Israel. Lancet. 2017;389(10088):2542-2550. doi:10.1016/S0140-6736(17)30789-4

Payne

Laporte

Deber

Coyte

PC.

Counting backward to health care’s future: using time-to-death modeling to identify changes in end-of-life morbidity and the impact of aging on health care expenditures. Milbank Q. 2007;85(2):213-257. doi:10.1111/j.1468-0009.2007.00485.x

Marshall

McGarry

Skinner

JS.

The risk of out-of-pocket health care expenditure at end of life. Springer; 2010.

10.

Curtis

Engelberg

Bensink

Ramsey

SD.

End-of-life care in the intensive care unit: can we simultaneously increase quality and reduce costs?

Am J Respir Crit Care Med. 2012;186(7):587-592. doi:10.1164/rccm.201206-1020CP

11.

Meerding

Bonneux

Polder

Koopmanschap

van der Maas

PJ.

Demographic and epidemiological determinants of healthcare costs in Netherlands: cost of illness study. Br Med J. 1998;317(7151):111-115. doi:10.1136/bmj.317.7151.111

12.

Lubitz

Beebe

Baker

Longevity and Medicare expenditures. N Eng J Med. 1995;332(15):999-1003. doi:10.1056/NEJM199504133321506

13.

Yang

Norton

Stearns

SC.

Longevity and health care expenditures the real reasons older people spend more. J Gerontol B Psychol Sci Soc Sci. 2003;58(1):S2-S10. doi:10.1093/geronb/58.1.S2

14.

Zweifel

Felder

Meiers

Ageing of population and health care expenditure: a red herring?

Health Econ. 1999;8(6):485-496. doi:10.1002/(SICI)1099-1050(199909)8:6<485::AID-HEC461>3.0.CO;2-4

15.

O’Neill

Groom

Avery

Boot

Thornhill

Age and proximity to death as predictors of GP care costs: results from a study of nursing home patients. Health Econ. 2000;9(8):733-738. doi:10.1002/1099-1050(200012)9:8<733::AID-HEC564>3.0.CO;2-U

16.

Shugarman

Bird

Schuster

Lynn

Age and gender differences in Medicare expenditures at the end of life for colorectal cancer decedents. J Women’s Health. 2007;16(2):214-227. doi:10.1089/jwh.2006.0012

17.

Shugarman

Bird

Schuster

Lynn

Age and gender differences in Medicare expenditures and service utilization at the end of life for lung cancer decedents. Women’s Health Issues. 2008;18(3):199-209. doi:10.1016/j.whi.2008.02.008

18.

Shugarman

Campbell

Bird

Gabel

Louis

Lynn

Differences in Medicare expenditures during the last 3 years of life. J Gen Intern Med. 2004;19(2):127-135. doi:10.1111/j.1525-1497.2004.30223.x

19.

Filc

Davidovich

Novack

Balicer

RD.

Is socioeconomic status associated with utilization of health care services in a single-payer universal health care system?

Int J Equity Health. 2014;13(1):115. doi:10.1186/s12939-014-0115-1

20.

Davidovitch

Filc

Novack

Balicer

RD.

Immigrating to a universal health care system: utilization of hospital services by immigrants in Israel. Health Place. 2013;20:13-18. doi:10.1016/j.healthplace.2012.11.005

21.

Gielen

Remacle

Mertens

Patterns of health care use and expenditure during the last 6 months of life in Belgium: differences between age categories in cancer and non-cancer patients. Health Policy. 2010;97(1):53-61. doi:10.1016/j.healthpol.2010.03.002

22.

Kelley

Ettner

Morrison

Wenger

Sarkisian

CA.

Determinants of medical expenditures in the last 6 months of life. Ann Intern Med. 2011;154(4):235-242. doi:10.7326/0003-4819-154-4-201102150-00004

23.

Fassbender

Fainsinger

Carson

Finegan

BA.

Cost trajectories at the end of life: the Canadian experience. J Pain Sympt Manage. 2009;38(1):75-80. doi:10.1016/j.jpainsymman.2009.04.007

24.

Polder

Barendregt

van Oers

Health care costs in the last year of life—the Dutch experience. Soc Sci Med. 2006;63(7):1720-1731. doi:10.1016/j.socscimed.2006.04.018

25.

, et al Determinants of total end-of-life health care costs of Medicare beneficiaries: a quantile regression forests analysis. J Gerontol A. 2022;77(5):1065-1071. doi:10.1093/gerona/glab176

26.

Morid

Kawamoto

Ault

Dorius

Abdelrahman

Supervised learning methods for predicting healthcare costs: systematic literature review and empirical evaluation. American Medical Informatics Association Annual Symposium Proceedings; vol.1312, 2017.

27.

Van Vliet

Lamers

LM.

The high costs of death: should health plans get higher payments when members die?

Med Care. 1998;36(10):1451-1460. doi:10.1097/00005650-199810000-00003

28.

Kocot

Ferrero

Shrestha

Dubas-Jakóbczyk

End-of-life expenditure on health care for the older population: a scoping review. Health Econ Rev. 2024;14(1):17. doi:10.1186/s13561-024-00493-8

29.

Babitsch

Gohl

von Lengerke

Re-revisiting Andersen’s behavioral model of health services use: a systematic review of studies from 1998-2011. Psychosoc Med. 2012;9:Doc11. doi:10.3205/psm000089

30.

Cohen-Mansfield

Skornick-Bouchbinder

Hoshen

Brill

The relationship between health services standardized costs and mortality is non-linear: results from a large HMO population. Health Policy. 2017;121(9):1008-1014. doi:10.1016/j.healthpol.2017.07.004

31.

Charlson

Pompei

Ales

MacKenzie

CR.

A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8

32.

Condelius

Edberg

A-K

Jakobsson

Hallberg

IR.

Hospital admissions among people 65+ related to multimorbidity, municipal and outpatient care. Arch Gerontol Geriatr. 2008;46(1):41-55. doi:10.1016/j.archger.2007.02.005

33.

Laux

Kuehlein

Rosemann

Szecsenyi

Co-and multimorbidity patterns in primary care based on episodes of care: results from the German CONTENT project. BMC Health Serv Res. 2008;8(1):14. doi:10.1186/1472-6963-8-14

34.

Baker

Shapiro

Schur

CL.

Health insurance and access to care for symptomatic conditions. Arch Intern Med. 2000;160(9):1269-1274. doi:10.1001/archinte.160.9.1269

35.

Monheit

Vistnes

JP.

Race/ethnicity and health insurance status: 1987 and 1996. Med Care Res Rev. 2000;57(4 Suppl.):11-35. doi:10.1177/1077558700057001S02

36.

Hafner-Eaton

Physician utilization disparities between the uninsured and insured: comparisons of the chronically ill, acutely ill, and well nonelderly populations. J Am Med Assoc. 1993;269:787-787. doi:10.1001/jama.1993.03500060087037

37.

Clarfield

Bergman

Kane

Fragmentation of care for frail older people—an international problem. Experience from three countries: Israel, Canada, and the United States. J Am Geriatr Soc. 2001;49(12):1714-1721. doi:10.1046/j.1532-5415.2001.49285.x

38.

Pope

Kautter

Ellis

, et al Risk adjustment of Medicare capitation payments using the CMS-HCC model. Health Care Fin Rev. 2004;25(4):119.

39.

Van Veen

Van Kleef

Van de Ven

Van Vliet

. Improving the prediction model used in risk equalization: cost and diagnostic information from multiple prior years. Eur J Health Econ. 2015;16:201-218. doi:10.1007/s10198-014-0567-7

40.

Panay

Baloian

Pino

Peñafiel

Sanson

Bersano

Predicting health care costs using evidence regression. Multidiscip Dig Pub Inst Proc. 2019;31(1):74. doi:10.3390/proceedings2019031074

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.05 MB

End-of-Life Health Costs Were Predicted Primarily by Prior Health Costs,and Secondarily by Temporal,Health and Demographic Factors

Abstract

Keywords

Introduction

Methods

Design, Setting, Materials, and Participants

Measurements

Demographics

Medical Data

Health Services Standardized Costs

Statistical Analytic Strategy

Results

Demographic and Background Characteristics

The Potency of ‘HSSC in Years Before the Final Year of Life’ as an Independent Variable in a Regression Predicting HSSC in the Final Year of Life (the Dependent Variable)

The Relationship Between the Explanatory Power of the Regression and the Time Interval Between Predictors (Independent Variables) and the Dependent Variable Period

Length of Predicting Period

Time Proximity of Predicted Period to Death

Length of the Period Pertaining to the Dependent Variable

Discussion

Conclusion

Supplemental Material

sj-docx-1-inq-10.1177_00469580251326315 – Supplemental material for End-of-Life Health Costs Were Predicted Primarily by Prior Health Costs, and Secondarily by Temporal, Health and Demographic Factors

Footnotes

Acknowledgements

Author Contributions

Data Availability

Declaration of Conflicting Interests

Funding

Ethical Approval and Informed Consent

ORCID iD

Supplemental Material

References

Supplementary Material