Trends of hospitalizations among patients with both cancer and dementia diagnoses in New York 2007–2017

Abstract

Background:

Cancer and dementia have often been studied in isolation. We aimed to examine the spatiotemporal trend of inpatient admissions with both cancer and dementia diagnoses.

Methods:

Using state-wide inpatient claims data, we identified all hospital admissions for patients aged ≥50 years with both cancer and dementia diagnoses in New York State, 2007–2017. We examined the spatiotemporal trend of the admission using a novel Bayesian hierarchical model adjusting for socioeconomic factor, as measured by Yost index.

Results:

Admissions with the presence of both cancer and dementia diagnoses represented 8.5% of all admissions with a cancer diagnosis, and the proportion increased from 7.1% in 2007 to 9.7% in 2017. The median admission rate was 3.5 (interquartile range: 2.2–5.2) hospitalizations per 1000 population aged ≥50 years, which increased from 2.9 in 2007 to 3.7 in 2017. The admission rate peaked first in 2010 followed by a smaller peak in 2014, before stabilizing at a level higher than the pre-2010 period. Taking into account the spatiotemporal heterogeneity, we found that hospitalizations among those with both cancer and dementia diagnoses were associated with a higher socioeconomic status (the posterior median relative risk for Yost index= 1.046 (95% credible interval: 1.033–1.058)).

Conclusions:

Hospitalizations of patients with both cancer and dementia increased over time. Cancer care providers and healthcare systems should be prepared to provide prevention and management strategies and engage in complex medical decision-making for this increasingly common patient population comprised of individuals with cancer and dementia.

1. Introduction

Both cancer and dementia are two common chronic diseases, which afflict older adults and exert substantial health and economic burden on the patient and their families and caregivers, as well as the healthcare system.^1–7 In the United States, the proportion of the population living with either cancer or with Alzheimer’s disease and related dementias is projected to increase.^8–15 Given these disease trends and the rapid aging population,¹⁶ more individuals will be living simultaneously with cancer and dementia. Such growth would undoubtedly increase the health and economic burden for patients, caregivers, and healthcare systems, as patients with cancer and dementia have high healthcare utilization, healthcare cost, and care burden.^7,17–20 Having both conditions likely complicates the treatment decision-making process and formal/informal care support systems, due in part to the diminished capacity in communication and decision-making among patients with dementia. Despite these concerns, literature focused on healthcare utilization among patients with cancer and dementia is scarce.

Typically studies of healthcare utilization separately investigate patients with either cancer or dementia, with few considering any overlap between the two conditions.²¹ We are aware of only one population-based study that examined the healthcare utilization among older adults with coexisting dementia and cancer, where the specific study population consisted of fee-for-service Medicare beneficiaries residing in the Mid-South region of the U.S. in 2009.²² However, trends of hospitalization among patients with both cancer and dementia diagnoses in other regions of the country and among patients with different health insurance coverage have not been examined. In addition, little is known about how the hospitalization admission rate for patients with both cancer and dementia has changed over space and time, and how these changes are affected by area-level socioeconomic factors.

To fill these research gaps, we provides a first estimate of the growing burden of dementia among cancer patients and survivors in a large population during an 11-year time span, using all payer hospital discharge data in New York State from 2007 to 2017. We focused on quantifying not only the time trend, but also the geographic variations in healthcare utilization. In addition, we assessed the impact of area-level socioeconomic factors on hospitalizations of those with both cancer and dementia diagnoses. We also provided descriptive and bivariate comparisons of patient-level characteristics and inpatient outcomes. The study adds to the small but growing body of literature focused on the intersection of cancer and dementia, as opposed to individually examining the growing cancer or dementia populations. Findings from the current study can provide policy-relevant insights into resource allocation and preparation for the growing aging population with both cancer and dementia.

2. Methods

2.1. Data Sources

We utilized inpatient data from the New York Statewide Planning and Research Cooperative System (SPARCS), an all-payer claim database for hospital discharge information.²³ This research was approved by our Institutional Review Board (STUDY-18–00947-CR001), which included approved access to limited patient health information (e.g. ZIP code) in this secondary data analysis of the deidentified data.

We searched the primary and all available secondary diagnoses using the International Classification of Diseases codes (Appendix eTable 1) to identify whether discharges were for patients with a cancer diagnosis and whether the same hospitalization also accompanied by a dementia diagnosis.^24,25 Out of all 14,647,152 hospitalizations between 2007 and 2017 for patients aged 50 years and older, we found 3,594,476 admissions with a cancer diagnosis. After excluding 160,958 admissions from non-New-York residents, the final analysis included 3,433,518 hospitalizations for patients with a cancer diagnosis.

2.2. Outcome variables

The main outcome of interest was hospitalizations among patients who had any diagnosis codes indicative of the presence of both cancer and dementia during the same admission. We summarized the absolute admission counts, and calculated the relative proportion of admissions wherein both cancer and dementia diagnoses were coded. We also calculated the admission rate defined as admission counts per 1000 population aged ≥50 years overall and according to calendar year and patient’s neighborhood of residence. Neighborhood was indicated by the ZIP Code Tabulation Areas (ZCTAs), which is an area unit defined by the US Census Bureau and is closely related to the postal ZIP codes.²⁶

2.3. Socioeconomic status index

We used the Yost index at ZCTA level as an indicator of the socioeconomic status (SES) of the patient’s residential neighborhood, a validated and widely used composite SES indicator in cancer disparity research.^27–30 It consists of the following SES attributes: occupation (proportion of working class), unemployment (proportion of population ≥16 years who are unemployed), poverty (proportion of population below 150% of poverty line), income (median household income), housing (median house value and median rent), and education (a weighted school years for the population ≥25 years). A higher Yost index indicates a better socioeconomic position. We calculated the ZCTA-level Yost index based on the method by Yu et al., 2014.²⁸

2.4. Statistical Analysis

Patient characteristics including inpatient outcomes were compared between those with both cancer and dementia diagnoses and with cancer but not dementia diagnosis using Chi-square test. Categorical variables for the number of comorbidities and the risk scores for in-hospital mortality and 30-day readmissions were based on calculations using commonly used algorithms.³¹ The algorithms use the diagnosis codes to identify clinical conditions that can affect mortality, hospital resource during the admission, and readmission.^32–34

We assessed the temporal trend of the proportion of admissions among patients with dual diagnoses of cancer and dementia by fitting a linear regression models treating admission year as a continuous variable. We additionally applied a segmented regression model to identify potential breakpoints of the temporal trend in admission rate.^35,36 Admission rates by the quartiles of Yost index were compared using Kruskal-Wallis test, and using Wilcox rank sum test for the paired comparison using the highest quartile as the reference.

The spatiotemporal pattern of admissions with both cancer and dementia diagnoses was examined using a novel Bayesian hierarchical model, which assessed both the overall temporal trend of admissions and the extent to which the spatial variation of admissions had changed over time ^37,38 This was implemented using Markov chain Monte Carlo simulation using the CARBayesST R package.³⁹ We ran three Markov chains, and each chain generated 220,000 samples, with the first 20,000 as burn-in period, and the remaining 200,000 were thinned with every 100th draw stored. The convergence was assessed using Geweke statistics. We fitted a Poisson regression model, given the nature of the count data of admissions, and used the total population ≥50 years as the offset. The 2010 Census ZCTA shapefile was used excluding three ZCTAs due to zero neighbors in the spatial weight matrix used in the model.⁴⁰ Missing values in ZCTAs with no admissions (5842/19701=29.7%), no population aged ≥50 years (594/19701=3.0%), or no Yost index (3398/19701=17.2%) were due in part to the mismatch between ZIP code and ZCTA, changes of ZCTAs across years, and unavailable estimates for calculating the Yost index.^26,29 To ameliorate the issue, we used the ZIP-to-ZCTA crosswalk files,⁴¹ and replaced with the corresponding minimums of admissions and population counts (1 and 0, respectively), and median Yost index (0.432). We ran a crude model without adjusting for any covariates, and an adjusted model with Yost index as the sole covariate.

All analyses were conducted using SAS (V9.4) and R (V4.0.2) with RStudio (V1.3.1073).

3. Results

3.1. Admission-level patient characteristics

Overall, 8.5% of all 3,433,518 admissions with a cancer diagnosis were also accompanied by a dementia diagnosis (Table 1). Of the patients with admissions coded with both cancer and dementia diagnoses, 55% were female, 47% age ≥85 years, 69% non-Hispanic whites, and 84% Medicare beneficiaries. Patients with both cancer and dementia differed statistically from those with cancer only for all characteristics examined (p<0.0001). Compared to patients with a cancer but no dementia diagnosis, patients with both cancer and dementia diagnoses were more likely to be 80 years or older, female, Medicare beneficiaries, had more comorbidities, lengthier hospitalizations, as well as higher risks of readmission and in-hospital death. They were also more likely to originate from and be discharged to SNF and less likely to be discharged home.

Table 1.

Characteristics of cancer hospitalizations with and without dementia among adults aged 50 years and over in New York State, 2007–2017.

Characteristicsa	Cancer without Dementia	Cancer with Dementiaa	All Cancer
Sample size n (%)	N= 3,141,814 (91.5%)	N= 291,704 (8.5%)	N= 3,433,518
Proportion	%	%	%
Sociodemographic
Sex
Male	47.64	44.97	47.41
Female	52.36	55.03	52.59
Age group (year)
50–64	31.41	4.17	29.09
65–69	14.15	3.98	13.28
70–74	14.23	7.79	13.69
75–79	14.06	14.01	14.05
80–84	12.59	22.62	13.44
85+	13.57	47.43	16.45
Race/Ethnicity
Non-Hispanic White	68.62	69.11	68.66
Non-Hispanic Black	13.52	14.2	13.58
Hispanics	7.71	8.01	7.73
Other	10.15	8.68	10.03
Insurance type
Medicare	59.63	84.06	61.7
Medicaid	5.01	2.03	4.75
Private	32.83	12.45	31.1
Self	1.97	1.11	1.9
Other	0.57	0.36	0.55
Comorbidities
Elixhauser comorbidity (n) b
0	10.42	4.44	9.91
1–2	44.4	38.4	43.89
3	19.66	22.98	19.94
≥4	25.52	34.18	26.25
Admission and discharge status
Admission originated from SNF	1.7	9.42	2.36
Discharge to SNF c	12.81	43.88	15.39
Discharge to Home c	79.03	45.44	76.24
Inpatient outcomes
Readmission risk score group b,c
−4 to < 2	25.29	14.8	24.4
2 ≤ to < 9	23.33	20.53	23.09
9 ≤ to <21	27.01	30.92	27.34
≥21	24.37	33.76	25.17
Mortality risk score group b,d
−25 to < −1	15.44	14.53	15.37
−1 ≤ to < 2	32.59	21.99	31.69
2 ≤ to < 10	24.6	26.95	24.8
≥ 10	27.37	36.53	28.15
Length of stay (days) d
1 to < 2	13.61	6.55	13.01
2 ≤ to < 4	28.32	21	27.7
4 ≤ to < 8	32.12	36.94	32.53
≥ 8	25.95	35.51	26.76
In-hospital mortality	5.46	7.76	5.66

Note:

^a,Chi-square test was used to compare the characteristics of cancer admission with and without dementia. All tests are statistically significant at p<0.0001. Admissions from the same patient at different time points were treated as separate and independent events.

^b,Of the 29 Elixhauser comorbidity types, we excluded four specific comorbidities (lymphoma, solid tumor, metastatic cancer, and neurological disorder) as they overlapped with the outcome of interest. Comorbidity types and the scores of readmission risk and mortality risk were calculated using algorithms from the Healthcare Cost and Utilization Project (https://www.hcup-us.ahrq.gov/tools_software.jsp).

^c,Among those discharged alive only.

^d,Categorized by quartiles.

3.2. Temporal trends in admissions and admission rates

The proportion of crude admission counts with both cancer and dementia diagnoses increased from 7.1% in 2007 to 9.7% in 2017 (Figure 1). This proportion increased linearly with time (beta coefficient=0.25, p<0.001). Similar trend was found across age groups (Appendix eFigure 1a), where the beta coefficient was 0.31 (p<0.001), 0.47 (p<0.001), 0.43 (p<0.001), 0.23 (p<0.001), 0.12 (p=0.008), and 0.35 (p<0.001), respectively, for 50–64, 65–69, 70–74, 75–79, 80–84, and ≥85 years, respectively. The increasing trend was only significant among patients aged 65 years with Medicare (beta=0.34, p<0.001), Medicaid (beta=0.41, p =0.002), or other insurance type (beta=1.2, p=0.0003) (eFigure1b). Among those uninsured, the proportion of admissions with cancer and dementia peaked during 2010–2012 (eFigure 1b).

Figure 1.

Distributions of the proportion of admissions with both cancer and dementia diagnoses to all admissions with a cancer diagnosis among adults aged 50 years and over in New York State, 2007–2017.

The overall median admission rate for hospitalizations with both cancer and dementia diagnoses was 3.5 (interquartile range (IQR): 2.2–5.2) hospitalizations per 1000 population aged ≥50 years, which increased from 2.9 (IQR: 1.8–4.6) to 3.7 (IQR: 2.4–5.5) hospitalizations per 1000 population aged ≥50 years from 2007 to 2017 (Figure 2 and eFigure 2a). The segmented regression model identified 2011 as the breakpoint (p<0.0001, eFigure 2b), and estimated that the slope of the regression for the period of 2007–2011 was 5.14 (95% CI: 3.70 – 6.59), which was different from that of 2011–2017 (β=−0.23, and 95% CI: −1.23 – 0.78).

Figure 2.

The median (inter quartile range) of the admission rate of hospitalizations with both cancer and dementia diagnoses per 1000 population ≥50 years in New York State, 2007–2017.

The median admission rate for patients with cancer and dementia was 3.45 (IQR 2.21–5.00), 3.07 (IQR 1.84–4.69), 3.12 ((IQR 1.96–4.52), and 3.85 (IQR 2.44–5.56) hospitalizations per 1000 population aged ≥50 years for the Yost quartiles, respectively (eFigure 2c). The admission rates differed significantly across the quartiles of Yost index (p<0.001), with significantly (p<0.001) lower rates in ZCTAs with low SES (Yost index ≤75th percentile) than in ZCTAs with high SES (Yost index >75^th percentile).

3.3. Spatiotemporal trends of admission rates

The spatiotemporal Bayesian hierarchical model revealed an overall increasing trend of admission rate of cancer hospitalization with dementia across New York State (Figure 3), with a first peak during 2010–2011 (>5 admissions per 1000 population aged ≥50 years) followed by a smaller peak in 2014–2015 (~ 4.5 admissions per 1000 population aged ≥50 years), before stabilizing at >4 admissions per 1000 population aged ≥50 years. In contrast, the level during the pre-2010 period was below 4 admissions per 1000 population aged ≥50 years. The observed trend took into account the variations in both time and space (with the consideration that the spatial variation of admissions changed over time), and adjusted for Yost index. Results from the crude model showed similar underlying temporal trends with larger variations (eFigure 4), suggesting that SES status indeed played a role in the spatial and temporal variations of the admission rate across New York. Taking into account the spatiotemporal heterogeneity, we found that admissions with both cancer and dementia diagnoses was associated with higher socioeconomic status (the posterior median relative risk for Yost index= 1.046 (95% credible interval: 1.034–1.059)).

Figure 3.

Posterior distributions of the average temporal trend of admission rate in hospitalizations with both cancer and dementia diagnoses. The observed trend took into account the variations in both time and space (with the consideration that the spatial variation of admissions changed over time), and adjusted for Yost index. The red lines show the posterior medians from three MCMC chains, and the black lines show the 95% credible interval.

The median admission rates with the dual cancer and dementia diagnoses, both observed (Figure 4) and modeled (eFigure 4b), varied substantially across all ZCTAs. High SES regions (Yost index≥75^th percentiles) tended to cluster around major metropolitan centers of Buffalo, Rochester, Syracuse, Albany, and New York City (eFigure 4c), where the rates of admissions with both cancer and dementia also tended to be high. We also observed a cluster of high admission rates in the northeastern New York, where the proportion of population aged ≥50 years was high (eFigure 4d), which was also evident from the map of the model residuals (eFigure 4e), and the map of admission rate for hospitalization with cancer but not dementia (eFigure 4f). In Long Island, areas with admission rates≥75^th percentile for hospitalizations with dual cancer and dementia diagnoses tended to occur in the middle of the island rather than the east end of the island with high proportion of population ≥50 years and high SES. Meanwhile, there were also pockets of ZCTAs with both high admission rates and low SES throughout the state.

Figure 4.

Spatial variations of the observed admission rate of hospitalizations with both cancer and dementia diagnoses per 1000 population ≥50 years. We mapped the median values aggregated from all ZIP Code Tabulation Areas (ZCTAs) in New York State from 2007 to 2017. The color scheme from light to dark corresponding to the following percentile breaks: minimum-5%, 5% to 25%, 25% – 50%, 50% – 75%, 75% – 95%, and 95%-maximum.

4. Discussion

While cancer and dementia are both considered diseases of aging, they have too often been studied in isolation. We observed an increase over time in the proportion of hospitalized patients with both cancer and dementia diagnoses and its spatial distribution varied with area-level socioeconomic status. The increasing trend was aligned with projected increase for cancer and dementia separately.^14,42 This observation has implications for treatment and cost of care delivery for patients with both conditions. Treating cancer patients with existing or newly diagnosed dementia may require different care management and delivery approaches as compared to cancer patients with other common comorbidities such as diabetes and hypertension.^43–45 We found that patients with dual cancer and dementia diagnoses were in general sicker and older with higher numbers of comorbidities, longer length of stay, and higher risk of dying during the hospitalization. For example, their in-hospital mortality (7.75%) was 37% higher than patients hospitalized with cancer diagnosis but without dementia (5.65%). The overall in-hospital mortality in our data was similar to those found in national samples (5.8%).⁴⁶ Compared to cancer patients without dementia, patients with dual cancer and dementia diagnoses were also more likely to come from and be discharged to SNF and less likely to return home to the community. Our results were consistent with findings from a previous study among the Medicare population,²² and review findings showing high prevalence of the dual-diagnosis of cancer and dementia in SNF settings.²¹ At the national level, studies have shown that inpatient stay among Medicare population (traditional and Medicare Advantage) ranged between 6.4 to 6.7 million during 2009–2017,⁴⁷ and the hospitalization rate for adults 65 and above were stable, with decrease seen among those 85 and older during 2000–2010.⁴⁸ The observed increasing trend in this study draw attention to unique challenges that may arise regarding effective decision-making discussions on complex treatment and disease management options with patients suffering from impaired cognitive and physical functioning.^{17,18,21,49,50} In addition, families and caregivers may experience increased financial strain and caregiving demands.^4,51,52

Our analysis also revealed interesting spatiotemporal patterns in hospitalizations among patients with both cancer and dementia diagnoses in relation to demographic and socioeconomic factors. The overall positive associations between the Yost index and the admission rate for patients with both cancer and dementia may in part reflect the age-structure of the neighborhood and the socioeconomic status of their residents. We observed a co-occurrence of high (above the 75th percentiles) admission rates in areas with high SES and high proportion of adults aged 50 years and over. Other factors, such as disparities in mortality and healthcare access could also help explain the observed association between SES and hospitalization. Studies focusing on cancer or dementia have separately shown that individuals living in areas with lower SES tend to undergo less preventive services, receive less screening and treatment, and experience higher mortality compared to those in higher SES areas.^53–58 Such disparities along the disease continuum could lead to more premature mortality in areas of low SES and high healthcare utilization in high SES areas. Thus, when using the identified spatial heterogeneities of the admission rate among patients with both cancer and dementia diagnoses to identify high-risk areas for targeted prevention strategies, caution must be taken to avoid inadvertently exacerbate the existing disparities. How to identify at risk individuals to improve population-level health and reduce prediction algorithm bias remain an active area of research.^59–61 Future in-depth studies on the extent to which the area-level SES affect inpatient utilization and the subsequent outcomes among patients with dual diagnosis of cancer and dementia are warranted.

Of note, our findings of peak admission rates among patients with both cancer and dementia diagnoses in 2010 and 2014 are consistent with two key policy events: the timing of the implementation of the Patient Protection and Affordable Care Act (ACA) in 2010 and Medicaid expansion under the ACA in 2014. Elsewhere, studies have found that both the ACA and its Medicaid expansion increased insurance coverage, access to and utilization of care services including preventive care such as cancer screening and high-quality hospitals for cancer-directed surgeries, particularly among people younger than 65 years old and/or low income patients.^62–69 However, variations and mixed results have also been reported with regard to care utilization such as hospitalization.^62,70 In addition, studies have observed an increase in hospitalization during the first year after Medicaid expansion but no change in the second year, indicating the “pent-up” demand phenomenon, and a potential overload of the oncology workforce and hospitals.^71,72 In these existing studies, New York was grouped with other states with the Medicaid expansion. Future in-depth studies on the direct impact of ACA and Medicaid expansion on our study population are warranted.

The study was subject to a few limitations. First, both cancer and dementia were based on diagnosis codes in claims data, which do not reflect the disease severity, and are prone to coding errors, particularly for diseases such as dementia that are hard to diagnose and are often under reported in claims.^3,25 The underreporting of dementia in claims data may imply a higher prevalence of patients with both cancer and dementia diagnoses than what was currently observed. The unknown timing of disease onset in claims data also prevented us from differentiating between cancer patients who underwent active treatment, whose hospitalizations were likely to occur within the first year following diagnosis,⁷³ and those who were long-term cancer survivors. In addition, reasons for transferring from and to SNF were unknown. Also, we cannot rule out that changes in coding practice overtime might partly contribute to the observed trend in hospitalizations among patients with both cancer and dementia. Second, we did not differentiate whether an admission was primarily due to cancer, dementia, or other reasons. However, this may be sufficient for the purpose of estimating the overall hospitalization utilization trend and assess healthcare resource demand for the growing patients with both cancer and dementia diagnoses. Furthermore, we adjusted for ZCTA-level Yost index in the model; while individual level SES variables would be more preferable over area-level SES, such information is not available in SPARCS. Finally, our New-York-based findings may not be generalizable to other regions or at the national level. Our results may serve as hypothesis generating purposes, and the identified locations and patterns of high cancer admissions with dementia deserve further investigations.

Despite the limitations of reliance on claims-based diagnoses, our data spanning 11 years of hospitalizations of cancer patients regardless of insurance type, shed light on the epidemiological patterns of an understudied, but large and growing subgroup, those with cancer and dementia. These cancer patients are older, with potentially more complex medical and non-medical needs than cancer patients without dementia. While further studies on cancer care for patients with these two diagnoses are warranted, cancer care providers and healthcare systems should be prepared to provide prevention and management strategies and engage in complex medical decision-making for an increasingly common patient population, individuals with cancer and dementia.