Abstract
INTRODUCTION
Mild cognitive impairment (MCI) is common in older adults, but the burden on patients and health systems is not well understood. We aimed to estimate the impact of MCI on healthcare utilization and costs.
METHODS
This was a matched cohort study in UK Biobank comparing healthcare costs and Alzheimer's disease (AD) dementia incidence rates in participants with MCI to propensity score‐matched participants without MCI.
RESULTS
Of 164,508 eligible participants, 6605(4%) had cognitive testing scores consistent with MCI. Ten‐year inpatient costs were 7.6% higher in MCI versus matched no‐MCI participants, while 6‐year primary care costs were 9.1% higher. Among MCI participants, AD dementia incidence rates were substantially higher than in non‐MCI participants (7.2 5‐year incidence rate ratio 95% CI: 3.3 to 15.7), and eventual AD dementia accrued higher additional inpatient costs (mean £20,199) over 10 years.
DISCUSSION
MCI is characterized by modestly higher healthcare utilization and costs. Subsequent AD dementia diagnosis was strongly associated with costs.
Highlights
Baseline cognitive tests identified individuals with all‐cause MCI in the UK Biobank.
We compared individuals with MCI to propensity score‐matched participants without MCI.
Inpatient costs were 7.6% higher over 10 years, and primary care costs were 9.1% higher over 6 years for participants with MCI.
AD dementia incidence rate ratio was 7.2 higher in participants with MCI.
Among MCI participants, eventual AD dementia was a key driver of costs resulting in higher inpatient costs (mean £20,199) over 10 years.
Keywords: Alzheimer's disease dementia incidence, healthcare costs, healthcare resource utilization, mild cognitive impairment to Alzheimer's progression, risk factors
1. BACKGROUND
Current projections estimate that 131 million people worldwide will be living with dementia by 2050, 1 with global societal costs reaching US $2 trillion by 2030. 1 Alzheimer's disease (AD) is the leading cause of dementia, with its prevalence rising alongside global aging trends. Diagnosing AD early when patients have mild cognitive impairment (MCI) remains challenging since patients often present only once cognitive decline has advanced. 2
Estimating the prevalence of all‐cause MCI in the general population is challenging. MCI, which can be transient as well as stable or progressive, is diagnosed based on a combination of history taking, clinical assessment, and cognitive testing. MCI due to AD is characterized by progressive cognitive impairment, 3 and hence, early diagnosis and recognition of MCI and its risk factors may allow for disease‐modifying therapies or other interventions (eg, cardiovascular risk modification, brain health interventions) to delay the onset of dementia. 4 , 5 , 6 Delaying dementia onset could result in substantial healthcare cost savings. 7 However, the differentiation in healthcare costs between individuals with and without MCI in the general population, along with the subsequent healthcare costs in those progressing to dementia, is less well studied. Previous studies lacked comprehensive large‐scale population‐based data capturing accurate cognitive assessments, health outcomes, and detailed data on healthcare costs. 8 , 9 , 10 The UK Biobank (UKB) could address these gaps through its extensive collection of multimodal data in a large cohort of middle‐aged adults. Follow‐up for the UKB cohort is now sufficient for long‐term outcomes such as dementia to be ascertained.
We aimed to use the UKB to identify a cohort of participants with cognitive test scores consistent with all‐cause MCI at enrollment (ie, independent of seeking healthcare for potential symptoms) to estimate healthcare utilization and cost burden in this population compared to matched individuals without MCI. As secondary aims we estimated the risk of progression to AD dementia along with healthcare utilization and costs stratified by eventual AD dementia diagnosis.
2. METHODS
2.1. Study population and design
The UKB, a large population‐based cohort study, recruited 502,394 participants aged 40 to 69 years in the United Kingdom (UK) between 2006 and 2010. Baseline assessment was conducted at 22 assessment centers in England, Scotland, and Wales and included self‐report questionnaires covering behavioral, medical, family history, sociodemographic, and other characteristics, along with physical examinations and biological sampling. Participants were followed up through data linkage to death registry and inpatient hospital records for all participants and to primary care records for a subset of participants. 11 Detailed information on the study design, data collection, linkage, informed consent, governance, and ethical statements was published previously. 11
Nearly all participants undertook the visual pairs matching and reaction time cognitive tests at baseline. Additionally, subsets of individuals completed assessments of numeric memory, prospective memory, and fluid intelligence (Figure S1). 11 Cognitive test data from selected tests was used to define the MCI population (Section 2.2). Participants were included if they had complete baseline cognitive assessment, age, and sex data, did not have a dementia diagnosis of any type at baseline (International Classification of Diseases [ICD‐10] codes F00, F01, F03, or G30), and had not withdrawn data linkage consent.
A retrospective cohort design was used to compare participants with MCI to a propensity score‐matched cohort without MCI at baseline (Section 2.6). Participants were followed from the initial assessment date at baseline until most recent secondary care record linkage, death, or end of study (December 31, 2019).
2.2. Ascertainment of MCI
The baseline cognitive assessment battery consisted of five tests, including pairs matching, reaction time, prospective memory, fluid intelligence, and numeric memory, the details of which were previously described. 12
A composite cognitive function measure was derived using a previously published methodology in the UKB. 12 , 13 Briefly, cognitive test score distributions were assessed, with reaction time log‐transformed due to its skewed distribution. To treat outliers, scores beyond three standard deviations of the mean were winsorized. Principal component analysis was utilized to derive a composite cognitive function score from the transformed test results. Based on eigenvalues and scree plots, the first principal component was selected as the best summary measure (44.4% of variance explained) and normalized to express the composite score in terms of a z‐score. In keeping with previous research and clinical practice, 14 MCI status was defined using a z‐score cut‐off of 1.96.
To determine MCI status while maximizing sample size, complete case analysis was performed on eligible participants having non‐missing scores for reaction time, prospective memory, and fluid intelligence tests (Figure S1).
2.3. Demographic and clinical variables
Baseline sociodemographic factors included age, family history of AD, and educational attainment obtained from self‐reports, as well as 33 comorbidities (Table S1). Comorbidities were determined using the UKB first occurrence data field, which combines data from primary care, inpatient care, death data, and self‐reports to provide the earliest date of diagnosis for health outcomes using three‐digit ICD‐10 codes. Baseline measurements of biochemical markers of inflammation (serum hsCRP) and glycemic control (serum HbA1c), genetic sequencing data for apolipoprotein E (APOE) genotype, and brain image‐derived hippocampal volume, which were measured in a subset of participants, were also included.
2.4. Healthcare use and costs
Healthcare resource utilization (HCRU) included secondary and primary care based on UKB‐linked inpatient hospital episode statistics and primary care data, respectively. Primary care use was measured by the number of consultations per year and number of unique medications prescribed per year. Secondary care HCRU was measured by number of hospital admissions per year. Prescriptions were grouped by British National Formulary (BNF) chapter. 15 Each unique prescription was then counted, defined by the chemical substance regardless of changes in dosage or product form. Primary care‐linked data were only available for a subset of participants and for up to 6 years after study entry. Although secondary care records are available for up to 12 years, those older than 10 years were excluded due to the impact of Covid‐19 on healthcare systems.
Costs of inpatient admission were calculated per hospitalization spell, spanning admission to discharge. Unit costs were derived by taking the activity‐weighted mean of all hospital‐reported costs across linked inpatient records for each healthcare resource group and activity type. 16 Secondary care costs were linked to 2020/2021 national reference costs. 17
RESEARCH IN CONTEXT
Systematic review: Authors searched PubMed and expert reports for publications on MCI‐related costs. Some studies exploring differentiation in healthcare costs between individuals with and without MCI found moderate differences in costs, while others found substantial differences. Previous studies lack comprehensive population‐based data capturing accurate cognitive assessments, health outcomes, and data on healthcare costs in this population. None were UK based.
Interpretation: Ten‐year inpatient costs were 7.6% higher in MCI versus matched participants without MCI; primary care costs were 9.1% higher over 6 years. AD dementia incidence rates were 7.2 times higher in MCI participants. Depression, apolipoprotein E (APOE) ε4/ε4 genotype, AD family history, and older age were associated with higher risk of progression. Among MCI participants, eventual AD dementia accrued higher additional 10‐year inpatient costs (mean £20,199).
Future directions: Our findings underscore the need for early detection and determination of the MCI etiology to delay or prevent the onset of dementia in people with MCI.
National Health Service unit costs for primary care HCRU, which are standardized across England, were derived from Personal Social Services Research Unit data. 18 Prescription costs were calculated using net ingredient costs (the cost at list price excluding the value added tax), obtained from the national report on the net ingredient cost of all prescriptions dispensed. 19
Supplementary Methods (Sections 1.1 and 1.2) provide full details of the HCRU and cost methodology.
2.5. Outcome ascertainment
The primary outcomes of annual HCRU and costs were assessed in the MCI population and propensity score‐matched comparators and a subset of the MCI population who developed AD dementia during follow‐up.
Secondary outcomes were AD dementia incidence and risk factors for progression to AD dementia in the MCI population.
AD dementia incidence was identified using diagnosis codes (ICD‐10 codes F00 or G30) in linked hospital inpatient records and UKB first occurrence data fields, which were mapped from primary care, inpatient admissions, self‐reports, and death records. 20 AD dementia diagnoses were validated in the UKB with >70% positive predictive value. 21
2.6. Analysis
Analysis was conducted in three stages. First, the incidence of AD dementia was estimated in participants with and without MCI. Then, analysis was conducted to identify risk factors associated with subsequent AD dementia in the MCI cohort. Second, primary and secondary HCRU and costs were estimated in those with MCI and propensity score‐matched comparator groups without MCI. Finally, the relative contributions of MCI and subsequent AD dementia to health costs were estimated in both primary and secondary care settings.
Kaplan‐Meier was used to compare AD dementia incidence in the with‐ and without‐MCI groups, treating loss to follow‐up and death as non‐informative censoring. AD dementia incidence rates and 95% confidence intervals (CIs) were generated at 1, 5, and 10 years after baseline. Aalen‐Johansen estimation, accounting for the competing mortality risk, was also conducted as a sensitivity analysis. A multivariable Cox proportional hazard model was used to identify predictors of MCI to AD dementia progression by estimating hazard ratios (HRs) and 95% CIs for candidate risk factors (Table S2). Due to potential diagnostic misclassification of AD dementia, Parkinson's disease was excluded as a predictor. 22 Hippocampal volume was also excluded due to limited brain imaging data in the MCI cohort. Continuous variables were input directly, while categorical variables were coded binarily (highest vs other risk, Table S2). Competing mortality risk was accounted for via a Fine‐Gray model. A sensitivity analysis was conducted, in which the Cox model for predictive factors of AD dementia was repeated on the full UKB population with cognitive function composite score as a continuous variable. The proportional hazards assumption was assessed via Schoenfeld residual plots for each of the modeled covariates and was found to have been satisfied.
To enable the analysis of secondary care costs and utilization, each participant with MCI that had complete sociodemographic, lifestyle, biomedical, and comorbidity data (Table S2) was propensity score matched on these characteristics to four cognitively unimpaired individuals. Similarly, to enable analysis of primary care costs and utilization, MCI participants with complete sociodemographic, comorbidity, primary care, and prescription data were separately propensity score matched to four cognitively unimpaired individuals. The propensity scores were derived via multivariable logistic models and included sex, age, family history of AD/dementia, sociodemographic deprivation, educational attainment, smoking status, alcohol intake, APOE genotype, serum high‐sensitivity C‐reactive protein, hemoglobin A1c, and various comorbidities (Table S1). Our approach to propensity score matching aligns with recommendations to reduce bias by including both potential confounders (variables that influence only the outcome) and true confounders (variables that influence both exposure and outcome) measured at baseline in the propensity score model. 23 MCI participants were matched to four individuals within a 0.2 standard deviation calliper width of the propensity score logit. 24 Covariate distribution balance before and after matching was checked using summary statistics. Balance was further assessed using absolute standardized mean differences (aSMD) (with aSMD < 0.1 indicating good balance 25 ) and density plots visualizing propensity score distribution overlap.
HCRU and cost differences between all‐cause MCI participants and matched no‐MCI participants and between all‐cause MCI participants with subsequent AD dementia and those who remained AD dementia‐free were calculated. HCRU was measured as the mean annual inpatient admissions, primary care consultations, and unique prescriptions over discrete follow‐up intervals (eg, 0 to 1 year, 1 to 2 years) using a person‐time approach censoring at death, exit, or study end, whichever was earliest.
Analyses were conducted using R Statistical Software version 4.1.1 (R Foundation for Statistical Computing, Vienna, Austria), with MatchIt and cobalt packages for propensity score matching. The data were accessed through UKB (Application No. 90906). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were adopted (checklist is provided in the Supplementary Appendix).
3. RESULTS
Of 502,394 UKB participants, 164,508 (33%) had complete baseline cognitive assessments and study inclusion criteria (Figure 1, Supplementary methods 1.3). Of these, 6605 (4%) had cognitive testing scores consistent with all‐cause MCI, and 5419 of these had complete sociodemographic and comorbidity records and were eligible for further analysis. Cognitive test score distributions are shown in Figure S1.
FIGURE 1.
Study inclusion and matching flowchart.
The median age of the entire study cohort (N = 164,508) was 58 years (interquartile range [IQR]: 50 to 63) (Table 1) and follow‐up time ranged from 8 to 15 years (median 9.92) during which 541 (0.3%) UKB participants developed AD dementia. Compared to those never diagnosed with AD dementia (no‐AD dementia group hereafter), those developing AD dementia were more likely to have a family history of AD or dementia (25% vs 13%), educational attainment below secondary school level (29% vs 14%), and carrying at least one copy of the APOE ε4 allele (61% vs 25.2%; Table 1).
TABLE 1.
Baseline sociodemographic characteristics of MCI and UK Biobank participants with cognitive assessment.
| All eligible UK Biobank participants | ||||
|---|---|---|---|---|
| Characteristic | MCI cohort (complete data) | Overall | Never diagnosed with AD dementia | Eventual diagnosis of AD dementia |
| N | 5419 | 164,508 | 163,967 | 541 |
| Sex | ||||
| Female | 3108 (57%) | 89,653 (54%) | 89,381 (55%) | 272 (50%) |
| Male | 2311 (43%) | 74,855 (46%) | 74,586 (45%) | 269 (50%) |
| Median (IQR) age at baseline | 62 (55 to 65) | 58 (50 to 63) | 58 (50 to 63) | 66 (63 to 68) |
| Family history of AD or dementia | 589 (11%) | 21,057 (13%) | 20,920 (13%) | 137 (25%) |
| Educational attainment | ||||
| College or university degree | 888 (16%) | 56,346 (34%) | 56,229 (34%) | 117 (22%) |
| Other postsecondary education | 1171 (22%) | 38,356 (23%) | 38,229 (23%) | 127 (23%) |
| Secondary education | 1133 (21%) | 44,985 (27%) | 44,855 (27%) | 130 (24%) |
| None of the above | 2062 (38%) | 23,386 (14%) | 23,231 (14%) | 155 (29%) |
| Prefer not to answer | 165 (3.0%) | 1435 (0.9%) | 1423 (0.9%) | 12 (2.2%) |
| APOE genotype a | ||||
| Other | 824 (15%) | 24,277 (15%) | 24,247(15%) | 30 (5.5%) |
| ε3/ε3 | 3206 (59%) | 94,384 (59%) | 94,211 (59%) | 173 (33%) |
| ε3/ε4 | 1235 (23%) | 37,183 (23%) | 36,951 (23%) | 232 (44%) |
| ε4/ε4 | 144 (2.7%) | 3663 (2.3%) | 3576 (2.2%) | 87 (17%) |
| Unknown | 10 | 5001 | 4982 | 19 |
Abbreviations: AD, Alzheimer's disease; MCI, mild cognitive impairment.
Denominator for calculating these percentages is the number of participants with known APOE status in each column.
The propensity score matching process was successful for all participants with MCI, and no participants were excluded due to the caliper setting. There was strong evidence of covariate balance after matching the participants with MCI to those without MCI (Table S3 and Figures S2‐S3). Over 10 years, average inpatient costs more than doubled across all UKB participants from £574 to £1229 per person per year (ppy) (Table 2). Participants with MCI incurred 7.6% higher hospital costs versus matched no‐MCI participants, amounting to a £1001 per‐person cumulative difference at 10 years (Table 2). At baseline, annual costs were £866 versus £789 ppy for MCI and non‐MCI participants, respectively, rising to £1978 ppy versus £1669 ppy at 10 years after baseline. Similarly, higher hospital admissions occurred for those with MCI, increasing from 0.43 ppy at baseline to 0.82 ppy at 10 years, compared to 0.40 to 0.73 ppy for the no‐MCI cohort (Table 2). Annual inpatient costs and HCRU per person by calendar year showed similar upward trends over time, with higher costs and hospitalizations for the MCI participants versus no‐MCI participants (Figure S4).
TABLE 2.
Total hospital inpatient costs and admissions per person per year (means and standard deviations) since UK Biobank entry for the entire UK Biobank, participants with MCI, and matched participants without MCI.
| Years since UK Biobank entry | Entire UK Biobank (n = 502,394) | All‐cause MCI (complete cases) (n = 5419) | No MCI (matched) (n = 21,676) | Percentage difference in costs in MCI versus matched no‐MCI (%) | Cumulative difference in costs (£) in MCI versus matched no‐MCI | |||
|---|---|---|---|---|---|---|---|---|
| Admissions (SD) | Costs (£) (SD) | Admissions (SD) | Costs (£) (SD) | Admissions (SD) | Costs (£) (SD) | |||
| 0 | 0.29 (1.71) | 574 (2758) | 0.43 (2.43) | 866 (3205) | 0.40 (2.06) | 789 (3479) | 10 | 77 |
| 1 | 0.33 (1.88) | 677 (4111) | 0.54 (2.96) | 1040 (3955) | 0.45 (2.19) | 930 (4422) | 12 | 187 |
| 2 | 0.35 (1.98) | 728 (3750) | 0.56 (2.98) | 1227 (6510) | 0.48 (2.14) | 1023 (4620) | 20 | 391 |
| 3 | 0.38 (2.14) | 794 (3999) | 0.62 (3.00) | 1261 (5379) | 0.54 (2.52) | 1112 (4522) | 13 | 540 |
| 4 | 0.40 (2.13) | 846 (4093) | 0.61 (3.13) | 1216 (4268) | 0.54 (2.45) | 1199 (4859) | 1 | 557 |
| 5 | 0.43 (2.34) | 916 (4517) | 0.60 (2.74) | 1150 (3752) | 0.60 (2.66) | 1300 (5412) | −12 | 407 |
| 6 | 0.47 (2.52) | 993 (5441) | 0.71 (3.31) | 1551 (9474) | 0.65 (3.23) | 1368 (5926) | 13 | 590 |
| 7 | 0.49 (2.62) | 1047 (5458) | 0.75 (3.59) | 1607 (8030) | 0.69 (3.36) | 1509 (6437) | 6 | 688 |
| 8 | 0.52 (2.72) | 1116 (6503) | 0.74 (2.64) | 1628 (5796) | 0.75 (3.73) | 1691 (7955) | −4 | 625 |
| 9 | 0.55 (2.80) | 1194 (8151) | 0.83 (2.56) | 1857 (6336) | 0.79 (3.36) | 1790 (7159) | 4 | 693 |
| 10 | 0.55 (2.92) | 1229 (7561) | 0.82 (3.67) | 1978 (8890) | 0.73 (3.65) | 1669 (9096) | 19 | 1001 |
Abbreviations: MCI, mild cognitive impairment; SD, standard deviation.
Stratifying MCI and matched no‐MCI participants by subsequent AD dementia diagnosis demonstrated that subsequent AD dementia diagnosis was also associated with higher healthcare costs. Individuals diagnosed with AD dementia during follow‐up had approximately £20,000 higher cumulative healthcare costs per person over 10 years than those without regardless of MCI status (Table 3 and Figure 2).
TABLE 3.
Total hospital inpatient costs per person per year (means) since UK Biobank entry for participants with all‐cause MCI and a subsequent diagnosis of AD dementia (MCI + AD dementia) versus matched participants without MCI and without AD dementia (No MCI + No AD dementia) and participants with MCI and no subsequent AD dementia diagnosis (MCI + No AD dementia).
| MCI | No MCI | |||||||
|---|---|---|---|---|---|---|---|---|
| Years since UK Biobank entry | AD dementia (n = 75) costs (£) | No AD dementia (n = 5344) costs (£) | Percentage difference in costs in MCI + AD dementia versus MCI + No AD dementia (%) | Cumulative difference in costs in MCI + AD dementia versus MCI + No AD dementia | AD dementia (n = 159) costs (£) | No AD dementia (n = 21,517) costs (£) | Percentage difference in costs in MCI + AD dementia versus No MCI + No AD dementia (%) | Cumulative difference in costs in MCI + AD dementia versus No MCI + No AD dementia |
| 0 | 1469 | 858 | 71 | 611 | 983 | 788 | 86 | 681 |
| 1 | 1615 | 1032 | 57 | 1194 | 1321 | 928 | 74 | 1369 |
| 2 | 1568 | 1222 | 28 | 1540 | 1590 | 1018 | 54 | 1918 |
| 3 | 1824 | 1253 | 46 | 2110 | 1267 | 1111 | 64 | 2631 |
| 4 | 2533 | 1198 | 111 | 3445 | 1429 | 1197 | 112 | 3967 |
| 5 | 2265 | 1135 | 100 | 4575 | 2315 | 1292 | 75 | 4940 |
| 6 | 4358 | 1514 | 188 | 7420 | 3211 | 1354 | 222 | 7944 |
| 7 | 3365 | 1585 | 112 | 9200 | 3077 | 1497 | 125 | 9811 |
| 8 | 3976 | 1600 | 149 | 11,575 | 3359 | 1679 | 137 | 12,108 |
| 9 | 6071 | 1811 | 235 | 15,836 | 4664 | 1770 | 243 | 16,409 |
| 10 | 6297 | 1934 | 226 | 20,199 | 6914 | 1635 | 285 | 21,070 |
Abbreviations: AD, Alzheimer's disease; MCI, mild cognitive impairment.
FIGURE 2.
Average hospital inpatient costs per year since cohort entry for participants with MCI and matched participants without MCI, stratified by eventual AD dementia diagnosis. MCI, mild cognitive impairment; AD, Alzheimer's disease.
In the 40% of patients with available primary care data, primary care appointment and prescription costs were 9.1% higher for MCI participants versus matched controls, from £372 ppy versus £346 ppy at baseline, rising to £427 ppy versus £377 ppy respectively at 6 years. Similarly, when compared to the matched cohort, MCI participants had a larger proportion of prescriptions for most therapeutic medication classes (Figure S5 and Table S4), with the largest differences observed for “Central Nervous System,” “Nutrition and Blood,” and “Skin” therapeutic classes. Additionally, subsequent AD dementia diagnosis was associated with £1300 per person (50%) higher primary care costs over 6 years compared to those not developing AD dementia (Figure S6).
AD dementia incidence also differed between individuals with and without MCI. Compared to matched participants without MCI (n = 21,676), AD dementia incidence was higher in the MCI participants after 5 years of follow‐up (67.4 per 100,000 person‐years, 95% CI 41.3 to 104.2 vs 9.3 per 100,000, 95% CI 4.8 to 16.5, rate ratio 7.2) and after 10 years of follow‐up (116.9 per 100,000 person‐years, 95% CI 90.0 to 149.3 vs 50.8 per 100,000, 95% CI 41.8 to 61.2, rate ratio 2.3) (Table 4). Among MCI participants who progressed to AD dementia, the median time to diagnosis was 6.6 years.
TABLE 4.
Estimated 1‐, 5‐, and 10‐year AD dementia incidence rates for MCI versus no‐MCI unmatched and matched populations.
| AD dementia incidence rate per 100,000 person‐years (95% CI) | |||
|---|---|---|---|
| Population | At 1 year | At 5 years | At 10 years |
| MCI | 45.5 (9.4 to 132.9) | 67.5 (43.5 to 100.4) | 116.7 (92.2 to 145.8) |
| No‐MCI (unmatched) | 0.6 (0.0 to 3.5) | 7.1 (5.5 to 9.2) | 29.6 (27.0 to 32.4) |
| MCI: No‐MCI AD dementia incidence rate ratio | 71.8 a | 9.5 (5.8 to 15.6) | 3.9 (3.1 to 5.0) |
| MCI | 55.4 (11.4 to 162.0) | 67.4 (41.3 to 104.2) | 116.9 (90.0 to 149.3) |
| No‐MCI (matched) | 0 (0 to 17.0) | 9.3 (4.8 to 16.5) | 50.8 (41.8 to 61.2) |
| MCI: No‐MCI AD dementia incidence rate ratio | ∞ | 7.2 (3.3 to 15.7) | 2.3 (1.7 to 3.2) |
Abbreviations: AD dementia, Alzheimer's disease dementia; MCI, mild cognitive impairment.
Exact number of AD dementia events at one year suppressed as < 5, therefore it was not possible to estimate 95% CI.
In Cox proportional hazards models, factors associated with AD dementia incidence in the MCI cohort were depression (HR, 95% CI: 2.92, 1.56 to 5.45), APOE ε4/ε4 genotype (2.42, 1.45 to 4.02), family history of AD dementia (2.03, 1.13 to 3.66), and age (1.17 per year, 1.10 to 1.24) (Table S5). Estimates of AD dementia incidence and associations with baseline characteristics in the MCI cohort were consistent after accounting for death as a competing risk (Figures S7‐S8).
4. DISCUSSION
People in their fifth or sixth decade with reaction time, prospective memory, and fluid intelligence cognitive test scores suggestive of MCI incur modestly higher healthcare utilization and costs compared to those without MCI at 10‐year follow‐up, even after adjusting for background comorbidities and MCI risk factors. The reason for this is not clear but may be a manifestation of the dementia prodrome. These findings contribute to our understanding of MCI epidemiology and could support efforts to characterize people at higher risk of future healthcare use related to dementia. Healthcare costs per person in the UKB cohort increased year on year, likely due to population aging. Our study found 7.6% higher annual inpatient costs for MCI compared to cognitively unimpaired comparators, similar to a 7.5% difference reported in a US‐based cohort study. 8 We also observed 9.1% higher primary care and prescription costs for MCI individuals, while the US study found a 16.3% difference in ambulatory care, which included laboratory and imaging test costs. 8 We found that participants with MCI received more prescriptions over time, particularly for nutrition/blood, skin, musculoskeletal/joint, and central nervous system conditions. This aligns with studies linking MCI to coexisting physical frailty, poor nutrition, depression, 26 and polypharmacy. 27
The ∼£20,000 cost difference over 10 years between those developing AD dementia and those who did not, regardless of initial MCI status, highlights AD's significant economic impact. It is plausible that this is simply a manifestation of the natural history of AD from MCI to dementia, with healthcare utilization and costs increasing over time as individuals’ healthcare needs increase due to progressive decline in cognitive function. Participants without MCI at UKB entry but who were subsequently diagnosed with AD dementia may have been cognitively intact at UKB entry and only developed MCI and subsequent AD dementia after enrollment in the cohort. The slight difference in cumulative costs between the participants with AD dementia according to their prior MCI status may be due to differences in disease trajectory, but this would need to be explored in other studies.
We found an overall MCI prevalence of 4.0% across all ages, increasing to 5.3% in individuals aged 60 and above. This is broadly consistent with previous studies, which estimated MCI prevalence at 6.0% to 6.7% in those over age 60. 28 AD dementia incidence in this study was 0.2 times lower than published rates in England 29 for ages 65+ (10 years: 29.6 versus 159 per 100,000 person‐years). Possible reasons include the younger UKB entry age (median 58 years vs ∼75 years) and potential healthy volunteer selection bias. 30 Extrapolation of the more recent Cognitive Function and Ageing Study II (CFAS II) trends 31 shows ∼0.2 times lower incidence for CFAS II 55 to 59 year versus 70 to 74 year age groups (1.5 vs 7.5 per 1000 person‐years; Figure S9). Thus, UKB AD dementia incidence aligns with extrapolated published age‐specific dementia rates.
Other studies have also reported higher dementia incidence in MCI versus non‐MCI controls. 32 , 33 A cohort study of 534 adults aged 70+ reported a HR of 23.2 (95% CI: 14.4 to 37.2) for developing dementia over 5 years. 34 A meta‐analysis of 6713 participants with MCI across six studies 35 found relative risks of 13.8 (95% CI: 8.4 to 22.6) for all‐cause dementia and 8.9 (95% CI: 4.2 to 19.1) for AD dementia over approximately 6 years, similar to our findings. Older age, 34 , 36 APOE genotype, 34 , 36 , 37 , 38 and depression 34 , 39 have also been associated with MCI‐AD dementia progression.
4.1. Strengths and limitations
Detailed cognitive testing enabled an unbiased data‐driven approach to generate composite scores for identifying MCI, validated against standard tests. 13 Robust data linkage to healthcare and mortality records during the long follow‐up period identified AD dementia outcomes with high case ascertainment. 21 Although some cases may be missed due to incomplete primary care data, recent research suggests that the direction and strength of associations are likely to remain consistent even with additional data. 40 Rich phenotyping of the UKB allowed for robust adjustments via propensity score matching for healthcare use and cost analyses and adjusted multivariate models predicting disease progression. Nevertheless, there remains a potential for residual confounding due to poorly measured or unknown confounders. The same set of variables was used for propensity score matching when estimating both healthcare utilization and dementia incidence in relation to MCI, providing methodological consistency. However, factors influencing healthcare utilization may differ from those affecting dementia progression, which may not be fully captured in our analysis.
This study also had limitations. First, MCI was identified using cognitive scores at enrollment without information on symptom onset, cognition trajectory, or biomarker evidence of underlying AD pathology, which might be important for identifying populations at high risk of AD dementia. Second, distinguishing MCI from cognitively unimpaired patients is challenging due to disagreement between classification systems. 35 , 41 However, using a UKB validated approach 12 , 13 and widely used cognitive score thresholds, 14 we successfully identified participants at high risk of AD dementia. While our MCI threshold of >1.96 SD below the mean may have missed some milder MCI cases and potentially included a small number of participants with undiagnosed AD, it minimized the inclusion of individuals with normal cognitive variation. Indeed, as discussed earlier, the observed MCI prevalence aligns with published estimates. Third, our propensity score matching approach may have underestimated MCI‐associated costs by controlling for comorbidities that could be mediators rather than confounders. Fourth, selection bias may exist as UKB participants tend to be healthy volunteers who may not be representative of the UK population. 30 However, differences in risk factor–outcome associations between UKB and other representative studies suggests associations from the UKB are similar to the general population. 42 Finally, the relatively small (N = 541) number of AD dementia diagnoses may have led to chance findings or left the study underpowered to detect associations. Nevertheless, our associations are consistent in magnitude and direction with published studies.
We have identified a MCI population at increased risk of AD dementia, but there is scope for further refinement by combining cognitive assessments with well‐established AD biomarkers like cerebrospinal fluid amyloid beta and tau peptides, positron emission tomography imaging, or magnetic resonance imaging measures of atrophy. 43 Emerging protein biomarkers, such as glial fibrillary acidic protein, 44 have been found to predict a decade earlier an eventual dementia diagnosis. This has implications for early diagnosis and population screening approaches because focusing on a biomarker‐confirmed MCI population is likely to reveal even higher per‐capita costs, strengthening the economic argument for early therapeutic intervention.
4.2. Conclusion
All‐cause MCI is associated with modestly higher healthcare utilization and costs, potentially reflecting both MCI‐related needs and those of coexisting conditions. A subsequent AD dementia diagnosis was strongly associated with healthcare costs, which is particularly important for the MCI population as they have a much higher risk of AD progression compared to those without MCI. These findings underscore the need for early detection and determination of the MCI etiology to delay or prevent the onset of dementia in people with MCI.
AUTHOR CONTRIBUTIONS
Sophie Edwards, Dominic Trepel, Craig Ritchie, and Marc Evans: conceptualization, interpretation of data, writing – review and editing. Julie Hviid Hahn‐Pedersen: conceptualization, methodology, writing – review and editing. Mei Sum Chan: conceptualization, formal analysis, methodology, writing – review and editing. Benjamin D. Bray: conceptualization, methodology, data verification, writing – review and editing. Alice Clark, AM, and Christian Ahmad Wichmann: methodology, validation, writing – review and editing. Mei Sum Chan and Benjamin D. Bray confirm they had access to and verified the data. All authors critically reviewed and approved the final version of the manuscript and accept responsibility for submitting it for publication.
CONFLICT OF INTEREST STATEMENT
CR is CEO and founder of Scottish Brain Sciences and reports personal fees from Actinogen, Biogen, Cogstate, Eisai, Eli Lilly, Janssen Cilag, Merck, Novo Nordisk, Roche Diagnostics, and Signant Health outside of the submitted work. SE reports no conflicts of interest. DT is associate editor for Alzheimer's and Dementia. MSC and BCB were employed by and BCB is a partner in Health Analytics, Lane Clark & Peacock LLP. JHHP, AC, AM, and CAW were employed by Novo Nordisk, which has two ongoing phase 3 trials in early AD and are shareholders of Novo Nordisk A/S. ME reports personal fees from NN, BI, AstraZeneca, and Moderna outside of the submitted work. A medical writer contracted by the sponsor provided assistance in preparing the manuscript. Author disclosures are available in the supporting information.
ETHICS STATEMENT
UKB received approval from the National Information Governance Board for Health and Social Care and the National Health Service North West Centre for Research Ethics Committee (Ref: 11/NW/0382). UK Biobank's research ethics committee and Human Tissue Authority research tissue bank approvals mean that researchers wishing to use the resource do not need separate ethics approval (unless re‐contact with participants is required).
CONSENT STATEMENT
UK Biobank has ethical approval from the North West Multi‐centre Research Ethics Committee. At baseline, all participants gave informed consent using a signature‐capture device. All participants provided consent for follow‐up through linkage to their health‐related records.
Supporting information
Supporting Information
Supporting Information
Supporting Information
ACKNOWLEDGMENTS
Funded by Novo Nordisk (NN), which participated in the study design, analysis and interpretation of data, writing of the report, and decision to submit the article for publication.
Ritchie C, Trepel D, Edwards S, et al. The impact of mild cognitive impairment on healthcare utilization and costs: A UK Biobank study. Alzheimer's Dement. 2025;17:e70065. 10.1002/dad2.70065
Craig Ritchie and Dominic Trepel, Joint first authors.
REFERENCES
- 1. Alzheimer's Disease International , Wimo A, Ali GC, Guerchet M, et al. World Alzheimer Report 2015: The global impact of dementia: An analysis of prevalence, incidence, cost and trends. 2015. Sep 21 [cited 2023 Dec 1]; Available from: https://www.alzint.org/resource/world‐alzheimer‐report‐2015/
- 2. Lin PJ, Neumann PJ. The economics of mild cognitive impairment. Alzheimers Dement J Alzheimers Assoc. 2013;9(1):58‐62. [DOI] [PubMed] [Google Scholar]
- 3. Albert MS, DeKosky ST, Dickson D, et al. The diagnosis of mild cognitive impairment due to Alzheimer's disease: recommendations from the National Institute on Aging‐Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement. 2011;7(3):270‐279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C. Potential for primary prevention of Alzheimer's disease: an analysis of population‐based data. Lancet Neurol. 2014;13(8):788‐794. [DOI] [PubMed] [Google Scholar]
- 5. Crous‐Bou M, Minguillón C, Gramunt N, Molinuevo JL. Alzheimer's disease prevention: from risk factors to early intervention. Alzheimers Res Ther. 2017;9(1):71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Fink HA, Jutkowitz E, McCarten JR, et al. Pharmacologic Interventions to Prevent Cognitive Decline, Mild Cognitive Impairment, and Clinical Alzheimer‐Type Dementia. Ann Intern Med. 2018;168(1):39‐51. [DOI] [PubMed] [Google Scholar]
- 7. Wimo A, Winblad B. Pharmacoeconomics of mild cognitive impairment. Acta Neurol Scand. 2003;107(s179):94‐99. [DOI] [PubMed] [Google Scholar]
- 8. Leibson CL, Long KH, Ransom JE, et al. Direct medical costs and source of cost differences across the spectrum of cognitive decline: a population‐based study. Alzheimers Dement J Alzheimers Assoc. 2015;11(8):917‐932. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Zhu CW, Sano M, Ferris SH, Whitehouse PJ, Patterson MB, Aisen PS. Health‐Related Resource Use and Costs in Elderly Adults with and without Mild Cognitive Impairment. J Am Geriatr Soc. 2013;61(3):396‐402. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Luppa M, Heinrich S, Matschinger H, et al. Direct costs associated with mild cognitive impairment in primary care. Int J Geriatr Psychiatry. 2008;23(9):963‐971. [DOI] [PubMed] [Google Scholar]
- 11. Sudlow C, Gallacher J, Allen N, et al. UK Biobank: an Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLoS Med. 2015;12(3):e1001779. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Lyall DM, Cullen B, Allerhand M, et al. Cognitive Test Scores in UK Biobank: data Reduction in 480,416 Participants and Longitudinal Stability in 20,346 Participants. PLOS ONE. 2016;11(4):e0154222. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Fawns‐Ritchie C, Deary IJ. Reliability and validity of the UK Biobank cognitive tests. PLOS ONE. 2020;15(4):e0231627. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Bickel H, Mösch E, Seigerschmidt E, Siemen M, Förstl H. Prevalence and Persistence of Mild Cognitive Impairment among Elderly Patients in General Hospitals. Dement Geriatr Cogn Disord. 2006;21(4):242‐250. [DOI] [PubMed] [Google Scholar]
- 15. BNF content published by NICE [Internet]. 2023. [cited 2023 Dec 4]. Available from: https://bnf.nice.org.uk/
- 16. Street A, Dawson D. Costing hospital activity: the experience with healthcare resource groups in England. Eur J Health Econ. 2002;3(1):3‐9. [DOI] [PubMed] [Google Scholar]
- 17. NHS England . 2020/21 National Cost Collection Data Publication [Internet]. 2022. Available from: https://www.england.nhs.uk/publication/2020‐21‐national‐cost‐collection‐data‐publication/
- 18. Curtis LA, Burns A. Unit Costs of Health & Social Care 2020 [Internet]. PSSRU, University of Kent; 2020:185. [cited 2023 Nov 24]. Available from:. https://www.pssru.ac.uk/project‐pages/unit‐costs/unit‐costs‐2020/ [Google Scholar]
- 19. Prescription Cost Analysis—England 2019 | NHSBSA [Internet]. [cited 2023 Dec 4]. Available from: https://www.nhsbsa.nhs.uk/statistical‐collections/prescription‐cost‐analysis‐england/prescription‐cost‐analysis‐england‐2019
- 20. UK Biobank . Health‐related outcomes data [Internet]. 2022. [cited 2023 Nov 24]. Available from: https://www.ukbiobank.ac.uk/enable‐your‐research/about‐our‐data/health‐related‐outcomes‐data
- 21. Wilkinson T, Schnier C, Bush K, et al. Identifying dementia outcomes in UK Biobank: a validation study of primary care, hospital admissions and mortality data. Eur J Epidemiol. 2019;34(6):557‐565. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Hunter CA, Kirson NY, Desai U, Cummings AKG, Faries DE, Birnbaum HG. Medical costs of Alzheimer's disease misdiagnosis among US Medicare beneficiaries. Alzheimers Dement J Alzheimers Assoc. 2015;11(8):887‐895. [DOI] [PubMed] [Google Scholar]
- 23. Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivar Behav Res. 2011;46(3):399‐424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Austin PC. Optimal caliper widths for propensity‐score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat. 2011;10(2):150‐161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Nguyen TL, Collins GS, Spence J, et al. Double‐adjustment in propensity score matching analysis: choosing a threshold for considering residual imbalance. BMC Med Res Methodol. 2017;17(1):78. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Kwan RYC, Leung AYM, Yee A, Lau LT, Xu XY, Dai DLK. Cognitive Frailty and Its Association with Nutrition and Depression in Community‐Dwelling Older People. J Nutr Health Aging. 2019;23(10):943‐948. [DOI] [PubMed] [Google Scholar]
- 27. Chippa V, Roy K. Geriatric Cognitive Decline and Polypharmacy. StatPearls [Internet]. StatPearls Publishing; 2023. [cited 2023 Nov 30]. Available from:. http://www.ncbi.nlm.nih.gov/books/NBK574575/ [PubMed] [Google Scholar]
- 28. Dunne RA, Aarsland D, O'Brien JT, et al. Mild Cognitive Impairment: the Manchester consensus. Age Ageing. 2021;50(1):72‐80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Imfeld P, Brauchli Pernus YB, Jick SS, Meier CR. Epidemiology, Co‐Morbidities, and Medication Use of Patients with Alzheimer's Disease or Vascular Dementia in the UK. J Alzheimers Dis. 2013;35(3):565‐573. [DOI] [PubMed] [Google Scholar]
- 30. Fry A, Littlejohns TJ, Sudlow C, et al. Comparison of Sociodemographic and Health‐Related Characteristics of UK Biobank Participants With Those of the General Population. Am J Epidemiol. 2017;186(9):1026‐1034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Matthews FE, Stephan BCM, Robinson L, et al. A two decade dementia incidence comparison from the Cognitive Function and Ageing Studies I and II. Nat Commun. 2016;7(1):11398. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Campbell NL, Unverzagt F, LaMantia MA, Khan BA, Boustani MA. Risk Factors for the Progression of Mild Cognitive Impairment to Dementia. Clin Geriatr Med. 2013;29(4):873‐893. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Kaufman DM, ed. Chapter 7 ‐ Dementia. Clinical Neurology for Psychiatrists. 6th ed. W.B. Saunders; 2007:111‐156. [Internet]. [cited 2023 Nov 30]. Available from:. https://www.sciencedirect.com/science/article/pii/B9781416030744100074 [Google Scholar]
- 34. Roberts RO, Knopman DS, Mielke MM, et al. Higher risk of progression to dementia in mild cognitive impairment cases who revert to normal. Neurology. 2014;82(4):317‐325. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Mitchell AJ, Shiri‐Feshki M. Rate of progression of mild cognitive impairment to dementia – meta‐analysis of 41 robust inception cohort studies. Acta Psychiatr Scand. 2009;119(4):252‐265. [DOI] [PubMed] [Google Scholar]
- 36. Li JQ, Tan L, Wang HF, et al. Risk factors for predicting progression from mild cognitive impairment to Alzheimer's disease: a systematic review and meta‐analysis of cohort studies. J Neurol Neurosurg Psychiatry. 2016;87(5):476‐484. [DOI] [PubMed] [Google Scholar]
- 37. Steenland K, Zhao L, John SE, et al. A ‘Framingham‐like’ Algorithm for Predicting 4‐Year Risk of Progression to Amnestic Mild Cognitive Impairment or Alzheimer's Disease Using Multidomain Information. J Alzheimers Dis. 2018;63(4):1383‐1393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Elias‐Sonnenschein LS, Viechtbauer W, Ramakers IHGB, Verhey FRJ, Visser PJ. Predictive value of APOE‐ε4 allele for progression from MCI to AD‐type dementia: a meta‐analysis. J Neurol Neurosurg Psychiatry. 2011;82(10):1149‐1156. [DOI] [PubMed] [Google Scholar]
- 39. Cooper C, Sommerlad A, Lyketsos CG, Livingston G. Modifiable Predictors of Dementia in Mild Cognitive Impairment: a Systematic Review and Meta‐Analysis. Am J Psychiatry. 2015;172(4):323‐334. [DOI] [PubMed] [Google Scholar]
- 40. Clifton L, Liu X, Collister JA, Littlejohns TJ, Allen N, Hunter DJ. Assessing the importance of primary care diagnoses in the UK Biobank. Eur J Epidemiol. 2024;39(2):219‐229. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Petersen RC, Knopman DS, Boeve BF, et al. Mild Cognitive Impairment: ten Years Later. Arch Neurol. 2009;66(12):1447‐1455. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Batty GD, Gale CR, Kivimäki M, Deary IJ, Bell S. Comparison of risk factor associations in UK Biobank against representative, general population based studies with conventional response rates: prospective cohort study and individual participant meta‐analysis. BMJ. 2020;368:m131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Dubois B, von Arnim CAF, Burnie N, Bozeat S, Cummings J. Biomarkers in Alzheimer's disease: role in early and differential diagnosis and recognition of atypical variants. Alzheimers Res Ther. 2023;15(1):175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Guo Y, You J, Zhang Y, et al. Plasma proteomic profiles predict future dementia in healthy adults. Nat Aging. 2024;4(2):247‐260. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting Information
Supporting Information
Supporting Information