Abstract
Background: The growing burden associated with population aging, dementia and multimorbidity poses potential challenges for the sustainability of health systems worldwide. We sought to examine how the intersection among age, dementia and greater multimorbidity is associated with health care costs.
Methods: We did a retrospective population-based cohort study in Alberta, Canada, with adults aged 65 years and older between April 2003 and March 2017. We identified 31 morbidities using algorithms (30 algorithms were validated), which were applied to administrative health data, and assessed costs associated with hospital admission, provider billing, ambulatory care, medications and long-term care (LTC). Actual costs were used for provider billing and medications; estimated costs for inpatient and ambulatory patients were based on the Canadian Institute for Health Information’s resource intensive weights and Alberta’s cost of a standard hospital stay. Costs for LTC were based on an estimated average daily cost.
Results: There were 827 947 people in the cohort. Dementia was associated with higher mean annual total costs and individual mean component costs for almost all age categories and number of comorbidities categories (differences in total costs ranged from $27 598 to $54 171). Similarly, increasing number of morbidities was associated with higher mean total costs and component costs (differences in total costs ranged from $4597 to $10 655 per morbidity). Increasing age was associated with higher total costs for people with and without dementia, driven by increasing LTC costs (differences in LTC costs ranged from $115 to $9304 per age category). However, there were no consistent trends between age and non-LTC costs among people with dementia. When costs attributable to LTC were excluded, older age tended to be associated with lower costs among people with dementia (differences in non-LTC costs ranged from −$857 to −$7365 per age category).
Interpretation: Multimorbidity, older age and dementia were all associated with increased use of LTC and thus health care costs, but some costs among people with dementia decreased at older ages. These findings illustrate the complexity of projecting the economic consequences of the aging population, which must account for the interplay between multimorbidity and dementia.
The presence of multiple chronic conditions is termed multimorbidity1 and is associated with worse clinical outcomes than good health or the presence of a single chronic condition.2–5 Dementia is an important contributor to multimorbidity and factors that contribute to multimorbidity (such as vascular disease) can also cause dementia. Like multimorbidity, dementia increases in prevalence with age. Therefore, the aging of the general population is expected to lead to further increases in the burden of both multimorbidity and dementia. Since both dementia and multimorbidity are independently associated with increased health care costs and an increased likelihood of requiring long-term care (LTC),6,7 the intersection of age, dementia and multimorbidity poses potential challenges for the sustainability of health systems worldwide.8 Given the potential for overlap and statistical interactions between these various exposures, there is potential for bias if individual associations are used to examine associations with cost. Rather, an approach that simultaneously considers the associations among age, dementia, morbidity and health care costs is preferable.
We used a large population-based data set of all 827 947 people aged 65 years or older who lived in a defined geographic area to characterize the frequency of dementia and 30 other common chronic conditions. Our goal was to advance the literature by considering the interplay between these 3 key exposures and total health care costs.
Methods
Study design and participants
For this retrospective population-based cohort study, we assembled a cohort of adults aged 65 years or older who lived in Alberta, Canada, between April 2003 and March 2017. We followed participants from April 2003, their 65th birthday or their registration with Alberta Health (whichever was later) until March 2017, death or migration out of the province.
We report this study according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline.9
Data sources
We examined the associations among age, dementia and burden of morbidity with total health care costs, composed of costs related to hospital admission, provider billing, use of ambulatory or emergency care, medications and long-term care (defined as care and services for those who cannot live independently or who require on-site nursing care, 24-hour supervision or personal support).10
We used the Alberta Kidney Disease Network (AKDN) database, which incorporates administrative data from Alberta Health (the provincial health ministry) such as provider claims, hospital admissions and ambulatory care utilization; Alberta laboratory data and Alberta Blue Cross prescription data.11 All people registered with Alberta Health were included in the AKDN database; all Alberta residents are eligible for insurance coverage by Alberta Health and more than 99% participate in the program. We linked postal codes for the last known residential address of each participant to Statistics Canada’s Postal Code Conversion File Plus (www.statcan.ca) to obtain rural or urban status and neighbourhood (postal code) income quintiles for each relevant fiscal year.
Morbidities
We used a previously published list of validated algorithms for 29 chronic morbidities that could be applied to claims data and had positive predictive values of at least 70%:12 dementia, alcohol misuse, asthma, atrial fibrillation, lymphoma, non-metastatic cancer (breast, cervical, colorectal, pulmonary and prostate), metastatic cancer, chronic heart failure, chronic pain, chronic obstructive pulmonary disease, chronic hepatitis B, cirrhosis, severe constipation, depression, diabetes, epilepsy, hypertension, hypothyroidism, inflammatory bowel disease, irritable bowel syndrome, multiple sclerosis, myocardial infarction, Parkinson disease, peptic ulcer disease, peripheral vascular disease, psoriasis, rheumatoid arthritis, schizophrenia, and stroke or transient ischemic attack. Dementia was 1 of the 29 morbidities and was defined by the presence of at least 1 hospital admission or 2 physician claims within 2 years (the International Classification of Diseases, Ninth Revision [ICD-9] codes 290, 294.1 and 331.2 or the International Statistical Classification of Diseases and Related Health Problems, 10th Revision [ICD-10] codes F00-F03, F05.1, G30 and G31.1).13
Subsequently, we found a validated algorithm for gout14 meeting the above criteria, so gout was additionally included in the final set of chronic morbidities. We also considered chronic kidney disease as a 31st morbidity that was defined by any of the following: mean annual estimated glomerular filtration rate (eGFR) less than 60 mL/min per 1.73 m2; a median annual presence of albuminuria (albumin to creatinine ratio ≥ 30 mg/g, protein to creatinine ratio ≥ 150 mg/g or dipstick proteinuria ≥ trace); 2 outpatient physician claims for dialysis; or 1 hospital admission or 1 outpatient claim for kidney transplantation.12
We classified each participant with respect to the presence or absence of dementia and the 30 other chronic morbidities for each fiscal year.15 If a participant developed a morbidity within a fiscal year or at any point previously (look-back extended as far as April 1994 where records were available), we classified the patient as having the morbidity. Detailed methods for classifying morbidity status and the specific algorithms used are found elsewhere.12
Costs and long-term care
The primary outcome was mean annual total health care costs; the cost components were hospital admission, provider visits (primary care or specialist care), ambulatory care (including emergency department visits), medications and LTC. For all hospital admissions and ambulatory care classification system (ACCS) charges between fiscal years 2004 and 2017, we used the Canadian Institute for Health Information’s resource intensity weights (RIWs) from the administrative data and Alberta’s cost of a standard hospital stay (CSHS).16 We used grouper codes for ACCS charges from fiscal years 2004 to 2010 and RIW and CSHS for the years thereafter. Costs for provider visits (inpatient and outpatient) were the actual amounts charged to Alberta Health Services; for physicians on the alternative payment program we based costs on the mean amounts charged by the other physicians. Medication costs were those listed with Alberta Blue Cross.
We measured time residing in an LTC home (e.g., nursing homes, auxiliary hospitals) and estimated costs on the basis of the average daily cost (Can$218.16; from Alberta Health) of all such homes in Alberta (individual-level data on the type of LTC were not available). We classified participants as residing in LTC if they were discharged to an LTC home after hospital admission or if we identified 2 provider claims at least 30 days apart for services provided in an LTC home; we deemed LTC to have begun on the earlier of the date of discharge and the date of the first claim, respectively. All costs are reported in Can$1000 units and are inflated to 2017 costs using the Consumer Price Index for all items in Canada. All data (demographics, morbidities and costs) were linked and organized by participant and fiscal year.
Statistical analysis
We did analyses with Stata MP 15·1 (www.stata.com) and reported baseline (first fiscal year within follow-up) descriptive statistics as counts and percentages, or medians and interquartile ranges, as appropriate. To examine the associations between dementia, increasing morbidity burden and age with cost outcomes, we used generalized linear models with a zero-inflated negative binomial distribution17 and a log link. A number of models were considered initially (i.e., mixed, generalized estimating equations, structural equations) but because of the distributions of the cost outcomes (excess zeros with long right tails) only the zero-inflated negative binomials models converged. We allowed for intraparticipant correlation to correct any nonindependence in our standard errors by using a clustered sandwich estimator; this allowed participants to contribute multiple fiscal years of cost data.
We regressed outcomes on dementia, the number of other (nondementia) morbidities (categorized as none, 1, 2, 3 and ≥ 4), age (categorized as 65–74, 75–84 and ≥ 85 yr), their 3-way interaction and all three 2-way interactions, as well as sex, rural or urban residence and the lowest neighbourhood (postal code) income quintile. The 2-way interaction terms allowed 1 variable to modify the association between another variable and the outcome either synergistically (greater than the sum of 2 individual main effects) or antagonistically (less than the sum of 2 individual main effects). The 3-way interaction term allowed 1 variable to modify the association between the combined effects of 2 other variables on the outcome, again either synergistically or antagonistically.
We allowed all covariates to be updated each fiscal year; an offset term was used to account for the log of partial and full years. Only rural or urban residence and the lowest neighbourhood income quintile had missing values; both of these were 7%. In the models, missing values were imputed with the most frequent category (i.e., imputed as urban and not the lowest neighbourhood income quintile).
We also did additional analyses that further examined the oldest age groups categorized as 85–89, 90–95 and 95 years and older. We determined independence of residuals from fitted values by examining plots of residuals versus fitted values. The threshold for statistical significance was set at 0.05. We reported marginal means and contrasts with 95% confidence intervals fixed at dementia, age and number of morbidities categories. We compared differences in means between those with dementia and those without, between adjacent age categories and between adjacent morbidity categories using Wald tests.
Ethics approval
The institutional review boards at the universities of Alberta (Pro00053469) and Calgary (REB16-1575) approved this study and waived the requirement for participants to provide consent because of the large sample size and the retrospective nature of the study. Data were deidentified.
Results
Participant flow is shown in Figure 1. There were 827 947 participants; median follow-up was 6.5 years (range 1 d to 14.0 yr; 2% of participants left the province before the end of follow-up). Twenty-six percent of participants died during follow-up. Participants could contribute follow-up data to more than 1 age category: 661 755 participants were followed while 65–74 years of age, 354 161 while 75–84 years and 158 538 while 85 years and older (Table 1). The percentage of participants with dementia increased with age, from 1.4% to 21.9%. The percentage of participants who were men decreased with age, from 49.8% to 37.3%. More participants living with dementia resided in a low-income neighbourhood than participants without dementia in the group aged 65–74 years, but this difference diminished with increasing age and was absent among those aged 85 years and older. Compared with participants without dementia, participants with dementia consistently had more morbidity across all age groups.
Marginal annual mean costs
Total costs over the whole cohort were distributed as follows: 33% LTC utilization, 32% hospital admission, 13% provider billing, 11% ambulatory or emergency care and 10% medications. Total marginal annual mean costs and the individual component costs were higher in almost all age categories and number of morbidities categories if dementia was present (Appendix 1, Table S1, available at www.cmajopen.ca/content/10/3/E577/suppl/DC1; Figure 2). The increases in total costs from dementia ranged from $27 598 to $54 171 within age categories and number of morbidities categories (Table 2). Similarly, mean total costs and the component costs increased in parallel with the number of morbidities, whether dementia was present or not (increases in total costs ranged from $4597 to $10 655 per morbidity). Mean total costs (driven by LTC costs) and LTC costs increased in parallel with age (increases in LTC costs ranged from $115 to $9304 per age category), but there were no consistent trends for other component costs with increasing age (the change in aggregated non-LTC costs ranged from −$7365 to $921 per age category).
Among people without dementia, mean annual costs for medications, ACCS charges and claims increased for most number of morbidities categories, from 65–74 years of age to 75–84 years of age (range from −$84 to $76, from −$28 to $23 and from −$44 to $155, respectively) and then decreased from 75–84 years of age to 85 years of age (range from −$227 to −$76, from −$381 to −$99 and from −$264 to −$46, respectively) (Appendix 1, Table S2, Figure S1). Hospital admissions and LTC costs increased in parallel with age (range from −$153 to $3248 and from $115 to $2561 per age category, respectively; Table 2).
In participants with dementia, LTC was the largest component of costs followed by hospital admissions, claims, ACCS charges and medications. Mean costs for LTC and mean total costs (driven by LTC costs) increased with age among participants with dementia (increases in LTC costs ranged from $7401 to $9304 per age category) (Table 2, Figure 2). When LTC costs were excluded, older age was associated with lower costs, and there was an inverse association between hospital admission, medication, ACCS charge and claim costs and increasing age (decreases in non-LTC costs ranged from −$857 to −$7365 per age category).
In a sensitivity analysis, we further divided participants aged 85 years and older into the following categories: 85–89 years, 90–94 years and 95 years and older. There were 144 134, 69 615 and 23 736 participants contributing follow-up to each age group, respectively. The results were similar to those for the whole study population (Table 3, Figure 3). Dementia was associated with higher mean total costs and component costs, and there was also an association between higher costs and the number of morbidities (Appendix 1, Table S3, Table S4). For people with and without dementia, there was an association between LTC costs (and thus total costs) and increasing age. However, after LTC costs were excluded, there was evidence of decreasing costs among older participants, with those aged greater than 95 years having the lowest non-LTC costs in most analyses (Table 3; Appendix 1, Figure S2).
Inspection of the burden of morbidity among the oldest participants (data not shown) demonstrated that morbidity was infrequent among those without dementia. For example, among participants aged greater than 95 years and without dementia, 56% of the participant-years were lived with 0 or 1 morbidity.
Interpretation
In this population-based study of more than 800 000 older adults treated in a universal health system, we found strong graded associations between multimorbidity, dementia and total health care costs. Older age was associated with significant increases in mean annual health care costs, driven by a strong association between older age and LTC utilization. As expected, dementia was associated with high health care costs that were driven by high utilization of LTC.
After LTC costs were excluded, the presence of dementia appeared to modify the relation between age and costs, such that older age was associated with increased non-LTC costs among those without dementia, but not necessarily among those with dementia. In fact, when LTC resource use was excluded, there was evidence that older age was associated with lower mean annual costs among people with dementia, including lower costs for hospital admissions, medications and ambulatory care.
These findings provide insight into how health care costs may change over time in parallel with the anticipated aging of the general population and therefore which interventions should be given the highest priority to mitigate the consequences of this demographic shift. First, to the extent that the costs associated with multimorbidity and with dementia were higher among people of older age, an increased prevalence of these conditions will exaggerate the economic consequences of the aging population, whereas interventions that prevent these conditions (or reduce their severity) will have the opposite effect. Second, since LTC is such an important driver of health care resource use, providing additional supports to enable older people to live independently rather than enter long-term care would probably yield economic benefits as well as improve quality of life. This may prove more difficult for adults with dementia and for those with physical morbidities. Third, we do not have data that directly explain the inverse association between costs and age among people with dementia once LTC costs were excluded. One possibility is survivorship effects, where those who survive to advanced age despite having dementia may have less morbidity and thus require less costly care. An alternative (not mutually exclusive) possibility is that provider attitudes or patient preferences mean that the care provided to people with dementia is less aggressive than that provided to those without dementia, leading to lower individual costs associated with hospital admissions, medications, emergency care and provider claims. Costs associated with acute care for older people with dementia might be reduced further if LTC homes enhance their ability to provide services that are currently restricted to hospitals, which would in turn require additional training and resources. The ongoing reviews of LTC that have been triggered by the COVID-19 pandemic may offer an opportunity to consider these issues in detail.18
Previous studies and a high-quality systematic review have demonstrated that multimorbidity (with or without dementia) is correlated with higher health care costs.7,19–23 Prior work also demonstrates that dementia is associated with increased costs, especially those due to hospital admission.6,24 However, most such studies have not been able to assess LTC utilization, which is an important contributor to total costs in older people (representing 33% of all health care costs in our cohort). Some prior studies have examined how multimorbidity is associated with a broad range of health care costs, including LTC and other forms of social care.25,26 Most of these studies have reached conclusions similar to ours but have not explored the intersection between dementia, multimorbidity and costs as we were able to do. We also assessed a broader range of morbidities than most prior studies of costs and dementia (or costs and multimorbidity), many of which focus on highly prevalent conditions such as vascular disease and diabetes.23 An exception is the 2012 Symphony study from the United Kingdom, which assessed the intersection between age, dementia and a broad panel of morbidities in a smaller population (1026 participants with dementia).27 Our findings are generally consistent with those of the Symphony study, which also showed that multimorbidity and dementia are more strongly correlated with cost than is age by itself.
Overall, the available evidence suggests that accurately projecting the economic consequences of the aging population is a complicated task and the interplay between morbidity and dementia as well as increasing age must be taken into account.
To gain additional insights about how the aging population may influence health care costs, future studies should combine the projected population structure in the coming decades with data examining the interplay between age, multimorbidity, dementia and costs, such as the findings presented herein. A more detailed examination of which conditions (or which clusters of conditions) account for most multimorbidity-related costs would probably improve the precision of these future studies. Out-of-pocket costs and opportunity costs (for unpaid labour) borne by caregivers and families should also be captured by such studies, since they probably account for a substantial proportion of the total economic burden associated with dementia. Finally, new methods for preventing dementia, attenuating multimorbidity and promoting independent living among older adults should be an extremely high priority for future research.
Our study has important strengths, including its rigorous analytical methods, our use of validated algorithms for ascertaining the presence of dementia and morbidity, and the large, geographically defined cohort.
Limitations
Our study has limitations that should be considered. Studies using administrative data will underestimate the true prevalence of dementia and other morbidities compared with those that use data acquired with a gold standard method such as a structured interview; because health care utilization increases with age, our focus on people aged 65 years and older should reduce the extent of such underestimation. The validated algorithm that we used to classify participants with dementia has a positive predictive value of 93% and a sensitivity of 67%.13 Therefore, our analysis will have misclassified some participants with respect to dementia. To the extent that such misclassification may have been random rather than systematic, this should have tended to bias our findings toward the null and thus should not have affected the observed associations between age, dementia and costs. Although we were able to extend further studies by including LTC utilization, we were not able to capture costs from outside the health sector, such as those associated with unpaid care (a major contributor to the societal costs of dementia28) or private sector care, and out-of-pocket costs for patients and families.29 The exact per-person cost of LTC was not available as we had access only to the average per diem cost; costs for patients with greater care requirements such as those with dementia were probably underestimated (and overestimated for those with lower care needs).
We did not have information on functional status, frailty, disabilities or the severity of morbidities, and thus we relied on the total count of morbidities, which is relatively crude, albeit independently associated with a broad range of clinical outcomes.2 More severe morbidity for a given morbidity count would be expected to increase the likelihood of LTC, which might in turn affect costs. Random misclassification of morbidity would be expected to bias toward the null without affecting our conclusions. However, if the clinical consequences of individual morbidities or morbidity count actually vary by age or dementia status, this may have affected our results, and this possibility requires further investigation. Finally, we studied people from a single Canadian province with data available only until March 2017; thus, our findings may not apply to other settings.
Conclusion
Multimorbidity and dementia were associated with higher mean annual health care costs. As expected, older age was associated with increased use of LTC and thus health care costs among people with and without dementia, and dementia was associated with substantially increased utilization of LTC. However, whereas older age was associated with higher costs of hospital admission, medications, acute care and provider claims among those without dementia, the converse was true among those with dementia, for whom there was an inverse association between older age and total costs once costs attributable to LTC were excluded. These findings suggest that the task of projecting the economic consequences of the aging population is complicated and must account for the interplay between morbidity and dementia as well as increasing age.
Acknowledgements
Ghenette Houston provided administrative support and Sophanny Tiv provided technical support at the University of Alberta.
Footnotes
Competing interests: Matthew James has received investigator-initiated grant funding from Amgen Canada. Scott Klarenbach is director of the Alberta Real World Evidence Consortium (University of Alberta, University of Calgary and Institute of Health Economics), which conducts investigator-initiated, industry-funded research unrelated to this work. No other competing interests were declared.
This article has been peer reviewed.
Contributors: Marcello Tonelli and Scott Klarenbach conceived the study. Marcello Tonelli, Natasha Wiebe and Scott Klarenbach designed the study and drafted the manuscript. Natasha Wiebe performed the statistical analyses. All authors made substantial contributions to developing the manuscript and revising it for important intellectual content, and all approved the final version. All authors agreed to act as guarantors for the work.
Funding: This research was supported by a Canadian Institutes of Health Research grant (FRN 143211) and a Leaders Opportunity Fund grant from the Canada Foundation for Innovation to Marcello Tonelli. Marcello Tonelli was supported by the University of Calgary’s David Freeze Chair in Health Research. Sharon Straus holds a Tier 1 Canada Research Chair in Knowledge Translation and Quality of Care. The funders had no role in the design or conduct of the study; the collection, management, analysis or interpretation of the data; the preparation, review or approval of the manuscript; or the decision to submit the manuscript for publication.
Data sharing: The authors are not able to make their data set available to other researchers because of their contractual arrangements with the provincial health ministry (Alberta Health), which is the data custodian. Researchers may make requests to obtain a similar data set at https://sporresources.researchalberta.ca.
Disclaimer: This study is based in part on data provided by Alberta Health and Alberta Health Services. The interpretation and conclusions contained herein are those of the researchers and do not represent the views of the Government of Alberta or Alberta Health Services. Neither the Government of Alberta nor Alberta Health or Alberta Health Services express any opinion in relation to this study.
Supplemental information: For reviewer comments and the original submission of this manuscript, please see www.cmajopen.ca/content/10/3/E577/suppl/DC1.
This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY-NC-ND 4.0) licence, which permits use, distribution and reproduction in any medium, provided that the original publication is properly cited, the use is noncommercial (i.e., research or educational use), and no modifications or adaptations are made. See: https://creativecommons.org/licenses/by-nc-nd/4.0/
References
- © 2022 CMA Impact Inc. or its licensors