Comprehensive descriptive analysis of large Alzheimer's disease patient cohorts

Abstract

Background

Precise estimates of the prevalence of Alzheimer's disease (AD), the distribution of demographic characteristics, comorbidities, treatment plans, insurance types, cost of treatment and survival probabilities at various time points are crucially important to advancing our understanding and for improving future AD research studies.

Objective

We analyzed two of the largest and high-quality medical databases, Oracle EHR Real-World Data and IQVIA. The results provide the most complete description of the AD patients in the US.

Methods

We present high-accuracy summary statistics of many important variables related to AD patients. Proportions, means and 95% confidence intervals were provided for all levels of the categorical and quantitative variables.

Results

We report high accuracy estimates of the overall survival probabilities for the first five years after initial diagnosis, drug treatments and patterns of use, demographics, insurance types, hospitalization duration, number of hospital visits, and a detailed list of comorbidities. We also report estimates of the annual total average cost of treatment per patient as well as itemized allocations for drugs, hospitalizations, surgery, and management costs.

Conclusions

We present the most complete, detailed and high-accuracy descriptive analysis of AD patients to date.

Introduction

Alzheimer's disease (AD) is a complex neurological disorder characterized by multifaceted pathophysiology. It is the most prevalent type of dementia that impacts over 6.5 million individuals in the United States (US).^1–4 The importance and complexity of AD have initiated and sustained multidisciplinary research endeavors with diverse areas of focus such as genetic risk factors,^5–7 environmental and lifestyle factors^8–10 and novel drug treatments.^11–13 Up until very recently, the treatment options for AD were only symptomatic treatments belonging to one of the two main categories of choline esterase inhibitors, such as donepezil and the anti NMDA receptor drug memantine. These treatments can be administered as single therapy or in combination therapy. ¹⁴ The implementation of a multiple drug therapy for AD patients is based on the rationale that these drugs target distinct dysfunctional circuitries and possess the potential to synergistically preserve cognitive function. Such a therapeutic approach would involve the administration of an acetylcholinesterase inhibitor, such as donepezil, in conjunction with an NMDA receptor antagonist, such as memantine. Moreover, it is worth noting that the combined use of donepezil and memantine is generally safe.^15,16 Recent study have found that this combined treatment significantly decreases the average yearly number of hospital and emergency department visits of AD patients as well as significantly increases their five-year survival probability compared to same outcomes for patients adhering to the most common three other treatments.^17,18

In this study, we examine new and detailed insights into AD patients’ demographic characteristics, comorbidities, drug treatments and patterns of use, insurance types, survival probabilities, hospitalization duration, number of hospital visits, as well as the total and itemized (drug, hospitalization, surgery, management) yearly cost of treatment variables. We analyzed two of the largest AD cohorts, consisting of 137,112 patients and 247,529 patients, obtained from two of the largest, high-quality databases, the Oracle Real-World Data^TM data and the IQVIA insurance claims database. This study design allows us to calculate the highest precision estimates of the characteristics described above. Furthermore, to the best of our knowledge, the patterns of medication use and the total and itemized cost measures for AD patients have never been studied before. We expect that this highly accurate and novel descriptive analysis will improve the design and the focus of many impactful future AD research endeavors. Our work will enable subsequent studies to focus on specific treatments and combinations of treatments, patient cohorts, survival periods, cost effectiveness, and quality of life improvement.

Methods

Data

The first dataset used for cost analysis was queried from a subset of the IQVIA database, a large and comprehensive healthcare data repository that contains detailed information on healthcare providers, patient diagnoses, prescription drugs, insurance claims, and medical devices. ¹⁹

Our second dataset was obtained from the Oracle Real-World Data (ORWD), one of the largest, high-quality research medical databases. ORWD has provided the data foundation for several medical breakthroughs using statistical and machine learning approaches.^20,21

IQVIA database

The IQVIA database includes medical insurance claims to providers within the US. We had access to a random sample consisting of 25% of the subjects present in the entire database. This subsample consisted of 23,390,878 distinct patients, 98,177 diagnostic codes, 174,475 unique drug codes, and 2,964,011,729 claims.

IQVIA Alzheimer's disease dataset

We only analyzed the cost of treatment from this database using a cohort of 247,529 AD patients diagnosed between 2006 and 2021. This analysis included the total and itemized (drug, hospitalization, surgery, management) annual cost of treatment per patient.

Oracle EHR real-world data

Oracle electronic health record (EHR) Real-World Data (OEHRRWD) is a large collection of de-identified electronic health records obtained from over 110 health systems in the US and as of 2022, the database contains over 100 million patients and 1.5 billion encounters. The database also contains conditions, medications, procedures, and lab tables of patients. Given the size and longitudinal structure of OEHRRWD, it provides an ideal platform for advanced statistical methods and machine learning methods with applications in all specialties.

Oracle EHR real-world data Alzheimer's disease dataset

We obtained complete data on AD patients with their first diagnosis between January 2016 and December 2022. We identified AD patients using the following International Classification of Diseases 10^th edition codes “G30: Alzheimer's disease, G30.0: Alzheimer's disease with early onset, G30.1: Alzheimer's disease with late onset, G30.8: Other Alzheimer's disease, and G30.9: Alzheimer's disease, unspecified” from the Oracle EHR Real-World Data (OEHRRWD) database. This produced a total cohort of 137,112 AD patients. We excluded patients with incorrect data format or missing data from the initial 137,112 selected subset, resulting in a total of 135,283 patients in our study (1.33% decrease in sample size). Key variables were derived using standardized operational definitions to ensure reproducibility and consistency across patients. Age at diagnosis was calculated as the patient's age, in years, at the first diagnosis date of AD, identified from encounter or condition tables using curated ICD-10 code sets. Comorbidities were defined as the presence of specified chronic conditions before, during, or after initial diagnosis of AD, also determined through ICD-10 coding. Medication exposure was defined as the period during which a patient was actively receiving treatment, based on the integration of medication orders, dispensations, and inpatient administrations within facility encounters. The insurance type was determined by the payer category recorded in the encounter table. Treatment was defined as the agent (or absence thereof) administered on or after the date of initial diagnosis. A treatment switch occurred when treatment with the initial agent was discontinued, with a subsequent initiation of a different agent. Combination therapy was identified when multiple agents were simultaneously administered to a patient. Death was defined using the patient's date of death in the Oracle Health EHR database, which is updated from hospital discharge records and clinical documentation. Only deaths with confirmed status codes were considered as events in the survival analysis. Hospital visits were defined as unique inpatient admissions, identified using encounter identifiers in the Oracle Health EHR. Length of stay was calculated as the difference in days between the admission and discharge timestamps.

Statistical measures

The presented descriptive analysis provides summary statistics and confidence intervals of both categorical and quantitative variables. For all categorical variables, age group, gender, payer, race, marital status, comorbidities, and medications, we reported frequency counts, percentages, and 95% confidence intervals for the percentages using the large sample approximation formula. For all quantitative variables, average days in hospital or ED per year, average hospital and ED visits per year, total and itemized cost (drug, hospital, surgery, management) of treatment per year between 2006 and 2021, we reported means and 95% CI for the means. These measures were calculated for each patient by dividing their total days in hospital or ED, total hospital and ED visits, and total itemized cost by the total follow-up time. All steps of these analyses were carried out using version 4.4.0 of the R statistical environment.

Results

Demographics

We categorized the age at diagnosis into five groups as follows: less than 46 years, 46 to 65 years, 66 to 75 years, 76 to 85 years, and 85 years and above. The corresponding percentages were 0.26%, 4.6%, 16.22%, 48.86% and 30.06% respectively. The gender distribution among AD patients was as follows: 61.7% female, 38.24% male, and 0.07% unknown. The majority of AD patients were Caucasian, accounting for 81.51% of the total. They were followed by patients with unknown races, making up 8.83% of the population. African Americans constituted 6.53% of the patients, while the remaining 3.13% consisted of individuals who were identified as American Indian or Alaska Native, Asian, Hispanic, or Pacific Islander. Among these AD patients, 40.34% were married, 7.33% were divorced, 43.53% were single, and the marital status of 8.8% was unknown. Detailed summary statistics of the patient demographic characteristics are shown in Table 1

Table 1.

Summary statistics of the oracle EHR real-world data AD patient demographic characteristics, comorbidities and medication use.

Variable name	N	p	95% CI
Age at Diagnosis
Less than 46	348	0.26	(0.23, 0.28)
46–65	6217	4.6	(4.48, 4.71)
66–75	21,943	16.22	(16.02, 16.42)
76–85	66,104	48.86	(48.6, 49.13)
85 and more	40,671	30.06	(29.82, 30.31)
Gender
Female	83,465	61.7	(61.44, 61.96)
Male	51,729	38.24	(37.98, 38.5)
Unknown	89	0.07	(0.05, 0.08)
Payer
Medicaid	2021	1.49	(1.43, 1.56)
Medicare	72,272	53.42	(53.16, 53.69)
Miscellaneous/Other	21,151	15.63	(15.44, 15.83)
Other Government Programs	825	0.61	(25.85, 26.32)
Private Health Insurance	3043	2.25	(0.57, 0.65)
Self-pay/Charity	687	0.51	(2.17, 2.33)
Unknown	35,284	26.08	(0.47, 0.55)
Race
African American	8828	6.53	(6.39, 6.66)
American Indian or Alaska Native	557	0.41	(0.38, 0.45)
Asian	2925	2.16	(2.08, 2.24)
Caucasian	110,271	81.51	(81.3, 81.72)
Hispanic	575	0.43	(0.39, 0.46)
Pacific Islander	191	0.14	(0.12, 0.16)
Unknown	11,939	8.83	(8.67, 8.98)
Marital Status
Divorced	9918	7.33	(7.19, 7.47)
Married	54,572	40.34	(40.08, 40.6)
Single	58,887	43.53	(43.26, 43.79)
Unknown	11,906	8.8	(8.65, 8.95)
Condition
Cerebral Infarction (I60-I69)*
Absence	109,348	80.83	(80.62, 81.04)
Presence	25,935	19.17	(18.96, 19.38)
Diabetes (E10-E13)*
Absence	114,377	84.55	(84.35, 84.74)
Presence	20,906	15.45	(15.26, 15.65)
Overweight And Obesity (E66)*
Absence	131,775	97.41	(97.32, 97.49)
Presence	3508	2.59	(2.51, 2.68)
Hypertensive Diseases (I10)*
Absence	113,048	83.56	(83.37, 83.76)
Presence	22,235	16.44	(16.24, 16.63)
Other Forms of Heart Disease (I3-I5)*
Absence	87,064	64.36	(64.1, 64.61)
Presence	48,219	35.64	(35.39, 35.9)
Acute Kidney Injury and Chronic Kidney Disease (N17-N19)*
Absence	91,619	67.72	(67.47, 67.97)
Presence	43,664	32.28	(32.03, 32.53)
Medications
No Treatment	49,942	36.92	(36.66, 37.17)
Donepezil	30,419	22.49	(22.26, 22.71)
Drug Switchers	28,110	20.78	(20.56, 20.99)
Memantine	11,602	8.58	(8.43, 8.73)
Donepezil & Memantine	9621	7.11	(6.97, 7.25)
Rivastigmine	2295	1.7	(1.63, 1.77)
Memantine & Rivastigmine	1058	0.78	(0.74, 0.83)
Donepezil-Memantine	743	0.55	(0.51, 0.59)
Galantamine	681	0.5	(0.47, 0.54)
Memantine & Galantamine	336	0.25	(0.22, 0.27)
Donepezil & Memantine & Donepezil-Memantine	195	0.14	(0.12, 0.16)
Other	281	0.21	(0.18, 0.23)
Total	135,283	100	−

*Designations in parenthesis are ICD-10 codes.

Comorbidities

The most commonly reported comorbidities among all patients were Other Forms of Heart Disease (35.64%), Hypertensive Diseases (16.44%), Cerebral Infarction (19.17%), Diabetes (15.45%), Acute Kidney Failure and Chronic Kidney Disease (32.28%), and Overweight and Obesity (2.59%). Detailed summary statistics of the patient comorbidities are shown in Table 1.

AD medications

We considered five most prevalent medications commonly used in the treatment of Alzheimer's disease and other forms of dementia: donepezil (Aricept), galantamine (Razadyne), memantine (Namenda), rivastigmine (Exelon), and Namzaric (combination medication that contains both donepezil and memantine). Over 36% of the AD patients received no treatment, donepezil was the most commonly prescribed drug (22.49%), followed by memantine (8.58%), donepezil & memantine (7.11%), and rivastigmine (1.7%). Additionally, 4.13% of AD patients received other medications. We also observed that 20.78% of patients switched their medications throughout our study observation time. Detailed summary statistics of patient AD medication use are shown in Table 1.

Among the 28,110 patients who altered their treatment regimen during the study, 4802 patients transitioned between donepezil and memantine. The most prevalent switching pattern was from donepezil to memantine, accounting for 26.13% of the cases, followed by a switch from memantine to donepezil, totaling 1240 patients (25.82%). Additionally, 18.1% of the patients switched from donepezil to memantine and subsequently back to donepezil, while 12.66% switched from memantine to donepezil and then back to memantine. The remaining patients chose various other switching patterns, as presented in Table 2.

Table 2.

Summary statistics of oracle EHR real-world data AD patient patterns of medication use.

Switchers between donepezil and memantine	N	p	95% CI
Donepezil-Memantine	1255	26.13	(25.9, 26.37)
Memantine-Donepezil	1240	25.82	(25.59, 26.06)
Donepezil-Memantine-Donepezil	869	18.1	(17.89, 18.3)
Memantine-Donepezil-Memantine	608	12.66	(12.48, 12.84)
Donepezil-Memantine-Donepezil-Memantine	229	4.77	(4.66, 4.88)
Memantine-Donepezil-Memantine-Donepezil	196	4.08	(3.98, 4.19)
Donepezil-Memantine-Donepezil-Memantine-Donepezil	134	2.79	(2.7, 2.88)
Memantine-Donepezil-Memantine-Donepezil-Memantine	114	2.37	(2.29, 2.46)
Other patterns*	157	3.27	(3.17,3.36)
Total	4802	100	−

*All other patterns have frequencies under 0.01.

Insurance type

Table 3 shows the summary statistics of the AD patient payer type. Overall, for 26.08% of the participants, the type of insurance was unknown, the most common type of insurance was Medicare (53.42%) followed by Miscellaneous/other (15.63%).

Table 3.

Summary statistics of the oracle EHR real-world data AD patient payer type.

Payer	N	p	95% CI
Medicaid	2021	1.49	(1.43, 1.56)
Medicare	72,272	53.42	(53.16, 53.69)
Miscellaneous/Other	21,151	15.63	(15.44, 15.83)
Unknown	35,284	26.08	(25.85, 26.32)
Other Government Programs	825	0.61	(0.57, 0.65)
Private Health Insurance	3043	2.25	(2.17, 2.33)
Self-pay/Charity	687	0.51	(0.47, 0.55)
Total	135,283	100	−

Survival

There was no available information on the patient's status regarding leaving or staying at the medical centers contributing information to the database. Thus, we had to carry out survival analysis without censoring. The results would be exact under the assumption of patients leaving the study cohort completely at random. We calculated the proportions of survivors from the first five years after initial diagnosis. These estimates were based on the different sample sizes because different cohorts can be followed for one, two, three, four and five years after initial diagnosis. These sample sizes were 103,072, 82,759, 59,565, 37,513, and 17,620, respectively. The estimated proportions of survivors are 0.925, 0.875, 0.837, 0.813, and 0.788, respectively. In the absence of censoring, this analysis is equivalent to the Kaplan-Meier survival curve estimation. The detailed summary statistics are shown in Table 4, and the survival curve with 95% CIs is presented in Figure 1.

Table 4.

Survival proportions of oracle EHR real-world data AD patient for the first five years since initial diagnosis.

Years since initial diagnosis	Sample size	Number of survivors	Number of deceased	Proportion of survivors	95% CI
1	103,072	95,336	7,736	0.925	(0.923, 0.927)
2	82,759	72,430	10,329	0.875	(0.872, 0.878)
3	59,565	49,876	9,689	0.837	(0.833, 0.842)
4	37,513	30,488	7,025	0.813	(0.808, 0.819)
5	17,620	13,875	3,745	0.787	(0.780, 0.797)

Figure 1.

Kaplan-meier estimator for the survival probability for the first five years since initial diagnosis.

The average annual length of hospital or Ed stay and the average number of hospital or ED visits

We also included metrics that reflect the average length of hospital or ED stay and the average number of such visits per year for AD patients. These are surrogate measures for the well-being and general health status of the patients. We only investigated patients who had service dates from 2016 to 2022. We removed patients with missing discharge dates because metrics such as the average length of stay and hospital visits cannot be calculated for these patients. Patients at any time point who were in the hospital for more than 60 days at a time were excluded because this could signify a patient being in hospice service. We also excluded patients who had a follow-up period of less than a year. This removed small sample bias in calculating the annual average number of hospital visits and average hospital and emergency department length of stay and produced a final cohort 110,381 distinct AD patients. Similarly, Table 5 displays the average length of stay and the number of hospital visits per study year. From this table, we can see a common trend that as their average length of stay in the hospital per year increases, so does their average number of hospital visits per year. We can also deduce that the greatest average lengths of stay in the hospital for each group were as follows: age greater than 86 at 2.75, gender unknown at 3.06, and African American at 3.66. In terms of the average number of hospital visits per study year, the corresponding groups with the highest numbers were as follows: age 86 and greater at 0.72, gender unknown at 0.76, and African American at 0.81.

Table 5.

Summary statistics of the oracle EHR real-world data AD patient average hospital visits per year and average days in hospital per year.

Variable	Average hospital or ED visits per year	95% CI	Average days in hospital or ED per year	95% CI
Gender
Female	0.67	(0.66,0.67)	2.48	(2.45,2.52)
Male	0.7	(0.69,0.71)	2.87	(2.82,2.92)
Unknown	0.76	(0.57,0.95)	3.06	(2.09,4.02)
Medication
No Treatment	0.64	(0.63,0.65)	2.63	(2.58,2.68)
Done/Mema	0.44	(0.39,0.49)	1.19	(1.03,1.35)
Donepezil	0.71	(0.7,0.73)	2.78	(2.71,2.84)
nbsp;nbsp;nbsp;Donepezil/Done/Mema	0.39	(0.28,0.48)	1.31	(0.87,1.76)
nbsp;nbsp;nbsp;Donepezil/Galantamine	0.69	(0.3,1.08)	2.03	(0.68,3.38)
Donepezil/Memantine	0.54	(0.53,0.56)	2.26	(2.18,2.35)
Switched	0.76	(0.74,0.77)	2.6	(2.54,2.67)
Other	0.68	(0.66,0.69)	2.67	(2.59,2.75)
Race
African American	0.81	(0.79,0.84)	3.66	(3.52,3.8)
American Indian or Alaska Native	0.8	(0.67,0.93)	3.1	(2.6,3.59)
Asian	0.58	(0.55,0.62)	2.32	(2.16,2.47)
Caucasian	0.69	(0.68,0.7)	2.59	(2.56,2.62)
Hispanic	0.6	(0.52,0.67)	2.11	(1.78,2.45)
Pacific Islander	0.67	(0.52,0.83)	2.14	(1.67,2.62)
Unknown	0.47	(0.46,0.49)	2.29	(2.2,2.38)
Marital Status
Divorced	0.78	(0.76,0.81)	3.13	(3.02,3.24)
Married	0.65	(0.64,0.66)	2.47	(2.42,2.52)
Single	0.74	(0.73,0.75)	2.86	(2.81,2.9)
Unknown	0.39	(0.37,0.4)	1.79	(1.71,1.86)
Payer
Medicaid	0.97	(0.9,1.03)	4.59	(4.02,5.17)
Medicare	0.94	(0.93,0.95)	3.45	(3.41,3.49)
Miscellaneous/Other	0.83	(0.82,0.84)	3.76	(3.67,3.84)
Other Government Programs	1	(0.92,1.08)	3.85	(3.41,4.28)
Private Health Insurance	0.8	(0.76,0.83)	3.49	(3.26,3.71)
Self-pay/Charity	0.54	(0.49,0.59)	1.9	(1.65,2.15)
Age
Less than 46	0.67	(0.53,0.8)	2.38	(1.62,3.14)
47–65	0.66	(0.62,0.69)	2.67	(2.52,2.83)
66–75	0.66	(0.64,0.67)	2.7	(2.62,2.79)
76–85	0.66	(0.65,0.67)	2.53	(2.49,2.57)
86 and more	0.72	(0.7,0.73)	2.75	(2.7,2.8)
Overall	0.68	(0.67, 0.69)	2.63	(2.60, 2.66)

Cost of treatment

We analyzed data from the comprehensive IQVIA insurance claims database which contained costs associated with AD treatments. The reported total cost of treatment and component cost of treatment were not adjusted for inflation. These numbers encompass all-cause expenses for AD patients. The average itemized costs for drugs, hospitalization, surgery and management as well as the total cost per patient showed an approximately linear increase from 2006 to 2016. Over this time period, the corresponding expenditures increased from $$2303.09 to $$3402.27, from $$5275.54 to $$8101.87, from $$1320.55 to $$1534.52, from $$1241.53 to $$2311.21, and from $$11,277.16 to $$16,933.10 respectively. There was a pronounced decrease in all costs between 2016 and 2017 likely caused by the implementation of the revisions of the Affordable Care Act (ACA) in 2014. There has been an approximately linear increase of the total and itemized costs from 2017 to 2021. Over this latter period, the expenditures increased from $$2970.44 to $$3630.79, from $$4045.06 to $$4936.89, from $$802.97 to $$898.23, from $$1336.97 to $$1672.98, and from $$9390.56 to $$10,839.82 respectively. Cost of treatment details for the 2006–2021 time period are shown in Table 6.

Table 6.

Summary statistics of the costs of treatment of AD patients.

Year	Number of patients	Average drug cost	95% CI	Average hospital cost	95% CI	Average cost surgery	95% CI	Average cost of management	95% CI	Average Cost Total	95% CI
2006	69,234	$2303.09	(2257.91,2348.27)	$5275.54	(5010.81,5540.26)	$1320.55	(1293.61,1347.49)	$1241.53	(1220.27,1262.8)	$11,277.16	(11,098.87, 11,455.45)
2007	74,731	$2700.47	(2656.09,2744.84)	$6382.27	(6091.49,6673.04)	$1193.47	(1170.24,1216.7)	$1343.46	(1320.05,1366.87)	$13,186.51	(12,988.33, 13,384.69)
2008	76,324	$2870.83	(2827.38,2914.27)	$6760.33	(6469.2,7051.45)	$1181.83	(1158.07,1205.59)	$1400.05	(1376.57,1423.52)	$13,696.20	(13,494.79, 13,897.61)
2009	71,688	$3067.96	(3018.43,3117.5)	$7926.31	(7516.37,8336.25)	$1292.12	(1264.92,1319.32)	$1576.88	(1550.64,1603.13)	$15,752.57	(15,486.9, 16,018.24)
2010	65,360	$3052.44	(2994.17,3110.72)	$8360.45	(7976.94,8743.96)	$1358.33	(1329.99,1386.68)	$1698.99	(1667.57,1730.4)	$16,384.00	(16,115.48, 16,652.52)
2011	63,266	$3012.58	(2956.52,3068.64)	$8193.70	(7827.07,8560.33)	$1380.37	(1349.42,1411.32)	$1728.45	(1698.37,1758.52)	$16,013.43	(15,751.65, 16,275.2)
2012	58,059	$2795.51	(2737.34,2853.69)	$7724.54	(7386.08,8063.01)	$1421.12	(1386.22,1456.02)	$1849.95	(1812.16,1887.73)	$15,652.18	(15,390.22, 15,914.14)
2013	51,092	$2941.21	(2863.35,3019.07)	$7581.80	(7204.21,7959.38)	$1513.55	(1475.38,1551.72)	$2056.94	(2003.86,2110.02)	$15,745.04	(15,452.26, 16,037.82)
2014	47,680	$3091.43	(3004.38,3178.48)	$7564.49	(7225.12,7903.86)	$1558.66	(1518.67,1598.64)	$2177.41	(2116.75,2238.08)	$16,002.79	(15,715.39, 16,290.19)
2015	42,977	$3313.29	(3148.35,3478.23)	$8505.55	(8095.61,8915.48)	$1583.10	(1537.58,1628.63)	$2298.41	(2241.37,2355.45)	$17,358.82	(16,989.52, 17,728.12)
2016	39,342	$3402.27	(3249.23,3555.3)	$8101.87	(7642.92,8560.82)	$1534.52	(1486.29,1582.74)	$2311.21	(2256.78,2365.64)	$16,933.10	(16,555.94, 17,310.27)
2017	34,105	$2970.44	(2862.42,3078.46)	$4045.06	(3743.53,4346.59)	$802.97	(772.23,833.71)	$1336.97	(1309.53,1364.42)	$9390.56	(9161.85,9619.26)
2018	28,143	$3209.65	(3073.31,3346)	$3873.73	(3589.13,4158.33)	$850.52	(808.13,892.91)	$1453.42	(1415.22,1491.61)	$9347.67	(9100.07,9595.27)
2019	26,141	$3265.54	(3102.9,3428.19)	$4383.94	(4052.3,4715.58)	$878.25	(838.39,918.12)	$1573.52	−15,051,642.04	$10,033.53	(9742.12, 10,324.93)
2020	22,305	$3213.87	(3044.5,3383.23)	$4319.47	(3964.95,4673.98)	$850.22	(803.28,897.15)	$1421.82	(1385.25,1458.39)	$9618.57	(9331.92,9905.22)
2021	18,442	$3630.79	(3355.42,3906.15)	$4936.89	(4546.67,5327.11)	$898.23	(849.2,947.26)	$1672.98	(1624.7,1721.26)	$10,839.82	(10,488.79, 11,190.85)

Furthermore, as shown above, the estimated average cost per year per AD patient in 2021 was $$10,839.82. According to the American Alzheimer's Association, there are an estimated 6.5 million people living in the US with AD and more than 50 million people worldwide. Thus, the estimated total cost of AD in 2021 is approximately 70.46 billion dollars. As the US population ages, the number of AD in the US is expected to double by 2040. The cost of treatment adjusted for inflation (2% per year) in 2040 is $$15,670.84. That yields an estimated total cost of treatment of AD patients of 203.72 billion dollars in 2040.

Discussion

In the US, prior studies reveal that AD imposes a growing economic burden and only modest gains in survival. Inflation-adjusted total costs roughly doubled from about $$160–$$215B in 2010 to ∼$$305B in 2020 (and ∼$$450B when valuing unpaid care), with most spending tied to long-term services and informal caregiving rather than acute medical care. ²² Unit prices for care settings have also climbed, e.g., 2019–2023 median costs rose ∼9.5%/year for home health aides and ∼7.4%/year for assisted living.^22,23 Regarding survival, it has been shown that average post-diagnosis life expectancy is between 4 and 8 years, varying by age at diagnosis.^24,25 Our study provides a unique, detailed exploratory survival and cost of treatment analysis of the largest previously unstudied cohort. We present a novel comprehensive analysis of various important AD patient characteristics such as demographics, comorbidities, medications and patterns of use, insurance types, total and itemized cost of treatment, and yearly survival probabilities. We attained exceptional precision of our summary statistics as we analyzed data queried from two of the largest, high-quality databases, the Oracle EHR Real-World Data (OEHRRWD) and the IQVIA Insurance Claims databases that contain 33,258,186 and 23,390,878 patients, respectively. Our corresponding AD datasets contained 137,112 and 247,529 subjects which are two of the largest AD research sample sizes to date. The large sizes of our cohorts ensure the high precision of our descriptive analysis.

We have reported several important summary statistics such as the most common demographic characteristics of AD patients which were single marital status, aged 76–85, female and Caucasian. This insight reveals the most vulnerable stratum of the population that can be the target for intervention and early screening. Furthermore, the majority of the patients were on Medicare and only 0.5% were self-paying. These results not only showed the degree of cost of treatment coverage in the US but also identified the most vulnerable group of patients that might not be able to afford the medical cost of treatment. Regarding the actual cost of treatment, we have identified piecewise linear trends for the total yearly average cost per patient as well as the drugs, hospitalizations, surgery, and management itemized costs. These findings reveal both the impact of the Affordable Care Act on healthcare cost reduction and provide a foundation for precise future cost forecasts. We found that the most common five drug treatments were None, Donepezil, Switchers, Memantine, and Donepezil & Memantine. A recent study has identified that the combined use of donepezil and memantine decreases the average number of visits to hospitals and emergency department visits per year compared to no drug users and donepezil and memantine single drug users. ¹⁸ The findings can help estimate the reduction of hospital and emergency department visits that can be expected if patients are switched from other treatments to the combined Donepezil & Memantine treatment.

We also provide survival probabilities for the first five years after initial diagnosis. These survival curves show the fastest decline in the first two years thus identifying the most critical period for intervention and treatment. We showed a unique analysis of the patterns of drug switching among the patients that did not adhere to one treatment. Almost all patterns included one or more switches between donepezil and memantine and vice versa.

The strength of our study is that it is based on analyses of two of the largest, high quality electronic health records (Oracle EHR Real World Data) and insurance claims (IQVIA) databases available to date. Moreover, these are both commercial databases with limited availability due to the high cost of access. Our descriptive analyses enable scientists to get previously unavailable high-precision data summaries and insights into important AD-related variables. Moreover, another strength of our study is the comprehensive range of variables we presented to describe AD patients. These include demographics, comorbidities, medication use, survival probabilities for the first five years following initial diagnosis, patterns of medication use, patient payer type, average hospital visits per year, average days in the hospital per year, as well as total treatment costs. During our frequent discussions with doctors, they have consistently emphasized that, in addition to modeling, they are highly interested in seeing extensive and accurate summary statistics. By leveraging detailed insights into demographics, comorbidities, medication patterns, hospitalization rates, and treatment costs from large-scale EHR and insurance claims databases, researchers can identify gaps in care for underrepresented populations, develop targeted interventions, and validate findings across diverse cohorts. Our approach also serves as a model for integrating diverse real-world datasets, enabling future studies to generate robust, generalizable insights into patient outcomes and healthcare resource utilization. This study advances AD research by providing the most detailed real-world descriptive evidence to date on survival, treatment patterns, comorbidities, insurance coverage, and costs. By integrating two large-scale EHR and claims datasets, it establishes high-precision benchmarks that can guide comparative effectiveness research, cost-effectiveness modeling, and policy planning. Importantly, the stratified analyses reveal disparities by payer, race, and demographics that highlight equity gaps in care. While the descriptive design and potential database biases limit causal interpretation, these findings offer a robust foundation for future studies using advanced causal inference and machine learning to test interventions, forecast costs, and address disparities in AD treatment and outcomes.

On the other hand, both the Oracle and IQVIA databases, although nationwide and based on large samples (101 and 22 million patients, respectively), do not perfectly represent the US population, as individuals who are underinsured or of low socioeconomic status are likely not included in these databases. However, this is a challenge that has yet to be fully addressed, and significant efforts are underway to collect data on these populations. Furthermore, by design our study did not include statistical machine learning or artificial intelligence methods that could be used to adjust for confounders, but such research efforts could be designed and fine-tuned by the information presented in our work. Future studies can build on our comprehensive descriptive analysis to improve understanding of optimal treatments and advantageous interventions for AD.