Comparing Lookback Periods to Ascertain Alzheimer’s Disease and Related Dementias

Abstract

Claims data are a valuable resource for studying Alzheimer’s disease and related dementias (ADRD). Alzheimer’s disease and related dementias is often identified using a list of claims codes and a fixed lookback period of 3 years of data. However, a 1-year lookback or an approach using all-available lookback data could be beneficial based on different research questions. Thus, the purpose of this study was to compare 1-year and all-available lookback approaches to ascertaining ADRD compared to the standard 3-year approach. Using a cohort of Veterans hospitalized for heart failure (N = 373, 897), our results suggested high agreement (93% or greater) between the lookback periods. The 1-year lookback period had lower sensitivity (60%) and underestimated the prevalence of ADRD. These results suggest that 1-year and all-available lookback periods are viable approaches when using claims data.

Keywords

alzheimer disease claims data diagnostic algorithms lookback periods veterans

Significance Statement

• Ascertainment of ADRD using a 1-year lookback or all-available lookback of claims data shows high agreement with a 3-year lookback period.

• The 1-year lookback has reduced sensitivity and prevalence compared to 3-year.

• The all-available lookback has reduced specificity compared to 3-year.

Introduction

Alzheimer’s Disease and Related Dementias (AD/ADRD) are the most common forms of dementia, resulting in $321 billion in annual direct medical costs in the United States.¹ In 2022, over 6.5 million Americans aged 65 and older were living with Alzheimer’s Disease, a number projected to surpass 13.8 million by 2060.^1,2 Alzheimer’s disease and related dementias is a neurodegenerative disorder characterized by cognitive decline and functional impairment in daily life which can progress from mild cognitive impairment to severe dementia.¹ Variability in clinical presentation paired with high rates of comorbid illness pose a challenge to early and accurate AD/ADRD diagnosis.^3-5 In fact, it has been reported that up to 50% of all dementia cases are not identified by primary care providers.^6-8 Further, when cognitive decline is detected, misdiagnosis is prevalent. An analysis of the National Alzheimer’s Coordinating Center (NACC) patient database from 2005-2010 produced an AD diagnosis sensitivity rate of 71-87% and a specificity rate of 44-71% using the National Institute of Neurological Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association (NINDS-ADRDA) as well as the National Institute on Aging (NIA)-Reagan criteria highlighting the limitations of current diagnostic practices.⁹ Misdiagnosis of AD has contributed to excess Medicare expenditures of between $9500-$14000 per patient per year.¹⁰ The projected growth of dementia patients paired with diagnostic challenges underscore the importance of accurate ADRD case identification to inform research and health policy moving forward. Cardiovascular disease represents one of the top concurrent medical comorbidities in ADRD.¹¹ Heart failure patients are of particular concern when studying ADRD, as a meta-analysis showed they have nearly a 60% higher risk of ADRD,¹² and a separate meta-analysis showed the prevalence of dementia in heart failure patients is up to 41%.¹³ Additionally, the prevalence of heart failure is projected to increase 46% between 2012 and 2030 which may further increase already surging incidence rates of ADRD.¹⁴ Thus, identifying ADRD in heart failure patients is imperative for adequate discharge planning of services, setting, follow-up and support.

Medicare and Medicaid claims data are a valuable resource for studying ADRD health care utilization within a population. In order to facilitate the study of ADRD and other chronic conditions at an epidemiological level, The Center for Medicare and Medicaid Services’ (CMS) Chronic Conditions Warehouse (CCW) developed algorithms for researchers to identify persons living with dementia through billing claims.^15-17 The CCW uses a 3-year “lookback” time ascertainment window to extract these diagnoses. This time window is generally useful for capturing diseases and syndromes such as ADRD that persist over time.

However, using a fixed lookback period complicates examination of links between ADRD and commonly comorbid disorders that may be psychiatric (i.e., depression)¹⁸ or medical (i.e., heart disease)¹² as these conditions may result in cognitive decline and therefore a misdiagnosis of cognitive impairment due to Alzheimer’s disease pathology. Such disorders tend to vary over time and can remit or occur episodically,¹⁹ and thus are typically identihfied through the medical record via a 1-or 2-year lookback period instead.²⁰ Understanding the validity of using a 1-year lookback period to capture the presence of ADRD, or the recency of the diagnosis, is therefore important to enable examination of relationships with important comorbidities using the same timescale at a population level.

When considering lookback periods, there may be trade-offs between a fixed lookback period versus using all-available lookback data. A fixed lookback period allows for the uniform assessment of covariates for all members of the cohort, but also relies on using a least common denominator of available data that has the potential to leave large amounts of archival data unused. Other research suggests using all lookback data (i.e., extended as far back as possible given the data source).²¹ The all-available lookback approach may result in less biased classifications than using a fixed lookback period (e.g., 1-year or 3-year), because fewer diagnoses will be “missed” by using the additional available data that is ignored using fixed lookback periods. This research has been done on simulated data and has yet to be applied to ADRD diagnoses.

Whereas recent research has compared various algorithms using 1- vs 3-year lookback periods of Medicare and Medicaid claims data to identify cases of ADRD,^22,23 research has yet to test the performance of the CCW diagnosis codes using a 1-year lookback or all-available lookback data. Thus, the purpose of this study was to compare lookback periods of the CMS CCW ADRD diagnosis codes. Using the standard 3-year lookback period as a reference standard, we examined the sensitivity, specificity, positive predictive value, negative predictive value, percent agreement, and prevalence for the CCW diagnosis codes using a 1-year lookback period and an all-available data lookback period. We hypothesized that, compared to the 3-year lookback period, the all-available lookback period would have high sensitivity and specificity, whereas the 1-year lookback period would have reduced sensitivity.

Methods

Participants

This study was done using secondary data analyses on a cohort of Veterans with a primary admission diagnosis of heart failure (HF) from October 1, 2011, to September 30, 2020. If participants had multiple HF-related admissions, one hospitalization was randomly selected for inclusion to minimize bias in selection of individuals at specific stages of disease progression. Information about demographics, medical and psychiatric comorbidity, prior health care utilization and mortality were obtained from Veterans Health Administration (VHA) electronic medical records. All study procedures were approved by the IRB at the Providence VA Medical Center.

Measures

ADRD Diagnoses

ADRD diagnoses (i.e., Alzheimer’s disease, Pick’s disease, presenile or vascular or frontotemporal dementia, cerebral or senile degeneration, dementia) were identified via ICD-9 or ICD-10 codes in the VHA medical record. The Chronic Conditions Warehouse (CCW) definition of ‘Alzheimer’s Disease and Related Disorders or Senile Dementia’ informed the ADRD codes included (See Appendix 1 for all codes used by the CCW algorithm).¹⁷

Sample Characteristics

Demographic data (i.e., sex, age, race/ethnicity) were collected through the VHA electronic medical records.

Assessment Approaches

Fixed

In this approach, all medical and pharmacy claims during the fixed baseline period were used to indicate the presence or absence of ADRD (See Appendix 1 for all claims data used). The fixed lookback windows ended one day before the index hospitalization. The 2 fixed lookback windows were 1-year and 3-year.

All-Available Data

ADRD was re-assessed by extending the original 3-year baseline periods for each binary covariate to capture all-available data. For some Veterans, this meant including information on ADRD occurring several years before cohort entry, although we continued to exclude the index date from ADRD assessment. See Figure 1 for a depiction of the fixed vs all-available lookback periods.

Figure 1.

Flowchart Depicting Lookback Periods for ADRD Codes. Assessment Periods Went Back 1-Year, 3-Year, and to the Earliest-Available Data Point for the 1-Year, 3-Year, and all-Available Lookback Approaches, Respectively.

Data Analyses

The CCW diagnosis codes were used to examine 1-year, 3-year, and all-available lookback periods to capture an ADRD diagnosis. We tabulated the frequencies of each diagnosis, then calculated sensitivity, specificity, positive predictive value, negative predictive value, agreement, and prevalence. We used the 3-year lookback period as our reference group as recommended by the CCW,¹⁷ because traditional gold standards such as a clinical assessment were not available with these data.

Results

373, 897 Veterans hospitalized with HF across 129 VA medical centers (mean [SD] ages 72 [11], 97% male, and 73% White) were included in analyses. When examining the amount of lookback data available for the all-available lookback period, the average was 3.34 years, SD = 3.43 years), the median was 2.27 years, with a maximum of 27.60 years. See Table 1 for a detailed description of the sample.

Table 1.

Sample characteristics.

Characteristic	Results
Characteristic	Mean (SD) or N (%)
ADRD diagnosis (1-year), N (%)	64,162 (17%)
ADRD diagnosis (3-year), N (%)	107,690 (29%)
ADRD diagnosis (all-available), N (%)	127,069 (34%)
Age at index admission, mean (SD)	72 (11)
Age group, N (%)
18-54	16,801 (5%)
55-64	72,674 (19%)
65-74	135,417 (36%)
75-84	86,251 (23%)
85+	62,748 (17%)
Sex, N (%)
Female	9543 (3%)
Male	364,341 (97%)
Race or ethnicity, N (%)
White	271,383 (73%)
Black	74,478 (20%)
Hispanic	26,562 (7%)
Other	462 (.1%)
Not stated	1012 (.3%)
Sum of elixhauser, mean (SD)	5.8 (3.0)
30-Day mortality, N (%)	44,348 (12%)
365-Day mortality, N (%)	127,526 (34%)

Algorithm Results

The 3-year lookback period identified 107,690 individuals with ADRD yielding a prevalence of 29%, whereas the 1-year lookback period identified only 64,162 individuals with ADRD cases (17%). The all-available lookback period identified all the 107,690 individuals already identified through the 3-year lookback, plus additional 19,379 individuals with ADRD that were not identified through the 3-year lookback algorithm. Results are tabulated in Table 2. Next, we examined the sensitivity, specificity, positive and negative predictive values, agreement, and prevalence of the 1-year and all-available lookback periods, compared to the 3-year lookback period. When using the 1-year lookback period, sensitivity was relatively low (60%), whereas specificity, positive and negative predictive values, and agreement were all relatively high (92% or above). The prevalence of ADRD was estimated at 17% through the 1-year lookback algorithm. The all-available lookback period results showed high sensitivity, specificity, positive and negative predictive values, and agreement with the 3-year lookback (all 88% or above). Prevalence was estimated at 34%. Table 2 reports all results and 95% confidence intervals.

Table 2.

Cases Identified by Look-Back Period, and Performance of Different Lookback Periods Using 3-Year Lookback Period as the Reference Standard (95% Confidence Interval Shown in Parentheses).

Lookback Period			ADRD Diagnosis (3 Year; N = 107,690)		No ADRD Diagnosis (3 year; N = 266,207)	χ² P-value
1-Year	ADRD (1 year; N = 64,162)		64,162 (60)		0 (0)	<.001
1-Year	Non-ADRD cases (1 year; N = 309,735)		43,528 (40)		266,207 (100)
All-available	ADRD cases (all-available; N = 127,069)		107,690 (100)		19,379 (7)	<.001
All-available	No ADRD diagnosis (all-available; N = 246,828)		0 (0)		246,828 (93)
Lookback period	Sensitivity	Specificity	Positive predictive value	Negative predictive value	Agreement	Prevalence
1-Year	.60 (.59, .60)	1.00 (1.00, 1.00)	1.00 (1.00, 1.00)	.92 (.92, .92)	.93 (.93, .93)	.17 (.17, .17)
All-available	1.00 (1.00, 1.00)	.93 (.93, .93)	.88 (.88, .88)	1.00 (1.00, 1.00)	.95 (.95, .95)	.34 (.34, .34)

Finally, we identified demographic and clinical characteristics of participants identified using the 3 different lookback periods. The results are shown in Table 3, and suggest persons in the 1-year prior algorithm were older, had more comorbidities, and higher rates of 30- and 365-day mortality than the 3-year or all years prior algorithm. Due to the dependency of the data, we did not calculate P-values comparing the characteristics of these 3 cohorts.

Table 3.

Sample Characteristics by Algorithm.

Characteristic	CCW 1-year prior N = 64,162 Mean (SD) or N (%)	CCW 3 years prior N = 107,690 Mean (SD) or N (%)	CCW all years prior N = 127,069 Mean (SD) or N (%)
Age at index admission, mean (SD)	79.4 (10.0)	77.2 (10.4)	76.9 (10.4)
Age group, N (%)
18-54	481 (.7%)	1398 (1.3%)	1722 (1.4%)
55-64	4353 (6.8%)	10345 (9.6%)	12649 (10.0%)
65-74	16457 (25.6%)	33841 (31.4%)	41175 (32.4%)
75-84	19523 (30.4%)	30826 (28.6%)	36322 (28.6%)
85+	23347 (36.4%)	31277 (29.0%)	35198 (27.7%)
Sex, N (%)
Female	1393 (2.2%)	2459 (2.3%)	2949 (2.3%)
Male	62768 (97.8%)	105225 (97.7%)	124113 (97.7%)
Race or ethnicity, N (%)
White	46483 (72.4%)	78583 (73.0%)	92772 (73.0%)
Black	12840 (20.0%)	21041 (19.5%)	24870 (19.6%)
Hispanic	72 (.1%)	140 (.1%)	152 (.1%)
Other	4704 (7.3%)	7839 (7.3%)	9178 (7.2%)
Not stated	63 (.1%)	87 (.1%)	97 (.1%)
Sum of elixhauser, mean (SD)	6.8 (3.2)	6.6 (3.1)	6.5 (3.1)
30-Day mortality, N (%)	13008 (20.3%)	18510 (17.2%)	20559 (16.2%)
365-Day mortality, N (%)	33962 (52.9%)	51362 (47.7%)	57890 (45.6%)

Discussion

The purpose of this study was to compare lookback periods when using the CCW diagnosis codes to ascertain ADRD. Our results suggest the 1-year lookback period has lower sensitivity and prevalence, but high agreement with the recommended 3-year lookback period.¹⁷ The all-available lookback period added about 4 months (.34 years) of additional lookback data on average, but extended as far as an additional 24 years of lookback data. Moreover, the all-available lookback has high sensitivity, specificity, and agreement with the 3-year lookback period. These results suggest that the 1-year lookback period likely underestimates ADRD cases when compared to the other 2 approaches, whereas the all-available lookback period is comparable and may surpass the 3-year lookback period in accuracy of ascertaining ADRD. However, future research using a true gold standard such as a clinical assessment and/or biomarker evidence (e.g., amyloid positivity on a positron emission tomography scan) is needed to fully characterize the accuracy of the all-available lookback period algorithm as the longer lookback may plausibly be associated with lower specificity.

For researchers who wish to examine associations between ADRD and other conditions that have shorter lookback periods than 3 years, the results of this study suggest this is a viable option given the high level of agreement between the 1-year and 3-year lookback periods. However, such studies should acknowledge that ADRD is likely to be underestimated and interpretations should be adjusted accordingly. Moreover, there are a large number of people (18%) in the all-available lookback approach that are not in the 3-year window period. It is difficult to determine what this means, since individuals who had a diagnosis 10 years ago but never since could be individuals who were misdiagnosed (e.g., experienced delirium that was coded as dementia), reverted to normal cognition,²⁴ or other possible reasons. Without the ability to reference a true gold standard diagnosis, it is difficult to determine how to best handle these situations. Moreover, the additional labor required to clean decades of claims data for possible situations of miscoding may not outweigh the observed change in results.

Limitations include the homogeneity of the sample, lack of a true gold reference standard diagnosis, and focusing solely on a cohort living with heart failure. Moreover, our analyses are nested since everyone identified in the 1-year algorithm will be identified in the 3-year algorithm, with a similar issue for the 3-year and all available algorithms. The data based on a predominantly male post heart failure hospitalization VA cohort is limited in generalizability to other populations. Veterans with heart failure who require hospitalization represent a particularly vulnerable group,^25,26 including higher risk of delirium which may confound a diagnosis of dementia, and have higher burden of comorbidity including atherosclerotic cardiovascular disease, that may put them at higher risk of dementia compared to Veterans without such risk factors.^12,27 As such, whereas our results should be replicated in more diverse samples, our results may well be indicative of findings in other populations which are at elevated risk for cognitive disorders.

The results of this study suggest using a 1-year or all-available lookback approach is a viable alternative to the 3-year lookback approach. However, using the 1-year or all-available lookback period comes with additional limitations that should be acknowledged. Clinically, the all-available approach may be capturing ADRD codes which occurred due to diagnostic errors that were not repeated on subsequent claims, or because ADRD was present but not reported on subsequent claims. These results underscore the importance of clinical evaluation for dementia including cognitive assessment and assessment of instrumental activities of daily living. Indeed, claims do not appear to provide information that is accurate enough for diagnostic certainty but could serve as a method of identifying members in a health system or health plan who are at risk of ADRD and who therefore warrant further evaluation.

Supplemental Material

Supplemental Material - Comparing Lookback Periods to Ascertain Alzheimer’s Disease and Related Dementias

Supplemental Material for Comparing Lookback Periods to Ascertain Alzheimer’s Disease and Related Dementias by Zachary J. Kunicki, Thomas Bayer, Lan Jiang, Melanie L. Bozzay, McKenzie J. Quinn, Alyssa N. De Vito, Sheina Emrani, Sebhat Erqou, John E. McGeary, Andrew R. Zullo, Matthew S. Duprey, Mriganka Singh, Jennifer M. Primack, Catherine M. Kelso, Wen-Chih Wu, and James L. Rudolph in American Journal of Alzheimer's Disease & Other Dementias®

Footnotes

Declaration of Conflicting Interests

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

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Zachary J. Kunicki

Thomas Bayer

Alyssa N. De Vito

Wen-Chih Wu

Data Availability Statement

Data were obtained through a use agreement with the U.S. Department of Veterans Affairs and cannot be made available publicly.

Supplemental Material

Supplemental material for this article is available online.

Appendix

References

Alzheimer’s Association . 2022 Alzheimer’s disease facts and figures. Alzheimer’s Dementia 2022;18(4):700-789.

Brookmeyer

Abdalla

Kawas

Corrada

. Forecasting the prevalence of preclinical and clinical Alzheimer’s disease in the United States. Alzheimer's Dementia. 2018;14(2):121-129.

Chin

Teodorczuk

Watson

. Dementia with Lewy bodies: Challenges in the diagnosis and management. Aust N Z J Psychiatr. 2019;53:291-303.

Porsteinsson

Isaacson

Knox

Sabbagh

Rubino

. Diagnosis of early Alzheimer’s disease: Clinical practice in 2021. The Journal of Prevention of Alzheimer's Disease. 2021;8(3):371-386.

Leyhe

Reynolds

Melcher

, et al. A common challenge in older adults: classification, overlap, and therapy of depression and dementia. Alzheimer's Dementia. 2017;13(1):59-71.

Connolly

Gaehl

Martin

Morris

Purandare

. Underdiagnosis of dementia in primary care: variations in the observed prevalence and comparisons to the expected prevalence. Aging Ment Health. 2011;15(8):978-984.

Bradford

Kunik

Schulz

Williams

Singh

. Missed and delayed diagnosis of dementia in primary care: Prevalence and contributing factors. Alzheimer Dis Assoc Disord. 2009;23(4):306-314.

Wilkins

Meisel

Depke

Williams

Edwards

. Dementia undiagnosed in poor older adults with functional impairment. J Am Geriatr Soc. 2007;55(11):1771-1776.

Beach

Monsell

Phillips

Kukull

. Accuracy of the clinical diagnosis of alzheimer disease at national institute on aging alzheimer disease centers, 2005–2010. J Neuropathol Exp Neurol. 2012;71(4):266-273.

10.

Hunter

Kirson

Desai

Cummings

AKG

Faries

Birnbaum

. Medical costs of Alzheimer’s disease misdiagnosis among US medicare beneficiaries. Alzheimer's Dementia. 2015;11(8):887-895.

11.

Wang

J-H

Y-J

Tee

. Medical comorbidity in Alzheimer’s disease: a nested case-control study. J Alzheim Dis. 2018;63:773-781.

12.

Wolters

Segufa

Darweesh

SKL

, et al. Coronary heart disease, heart failure, and the risk of dementia: A systematic review and meta-analysis. Alzheimer's Dementia. 2018;14(11):1493-1504.

13.

Yap

NLX

Kor

Teo

, et al. Prevalence and incidence of cognitive impairment and dementia in heart failure - a systematic review, meta-analysis and meta-regression. Hellenic J Cardiol. 2022;67:48-58.

14.

Virani

Alonso

Aparicio

, et al. Heart disease and stroke statistics-2021 update: A report from the American Heart Association. Circulation. 2021;143(8):CIR0000000000000950.

15.

Taylor

Fillenbaum

Ezell

. The accuracy of medicare claims data in identifying Alzheimer's disease. J Clin Epidemiol. 2002;55(9):929-937.

16.

Taylor

Jr. Østbye

Langa

Weir

Plassman

. The accuracy of Medicare claims as an epidemiological tool: the case of dementia revisited. J Alzheimers Dis. 2009;17(4):807-815.

17.

Chronic Conditions Warehouse . Alzheimer’s Disease and related disorders or senile dementia. CCW; 2022. https://www2.ccwdata.org/web/guest/condition-categories

18.

Bennett

Thomas

. Depression and dementia: cause, consequence or coincidence? Maturitas. 2014;79(2):184-190.

19.

American Psychiatric Association . Diagnostic and Statistical Manual of Mental Disorders: DSM-5. Arlington, VA: American Psychiatric Association; 2013.

20.

Chronic Conditions Warehouse . 27 CCW Chronic Conditions Algorithms. CCW; 2022. https://www2.ccwdata.org/documents/10280/19139421/ccw-chronic-condition-algorithms.pdf

21.

Brunelli

Gagne

Huybrechts

, et al. Estimation using all available covariate information versus a fixed look-back window for dichotomous covariates. Pharmacoepidemiol Drug Saf. 2013;22(5):542-550.

22.

Grodstein

Chang

C-H

Capuano

, et al. Identification of dementia in recent medicare claims data, compared with rigorous clinical assessments. J Gerontol: Series A. 2021;77(6):1272-1278.

23.

McCarthy

Chang

C-H

Tilton

Kabeto

Langa

Bynum

JPW

. Validation of claims algorithms to identify Alzheimer’s disease and related dementias. J Gerontol: Series A. 2021;77(6):1261-1271.

24.

Canevelli

Grande

Lacorte

, et al. Spontaneous reversion of mild cognitive impairment to normal cognition: A systematic review of literature and meta-analysis. J Am Med Dir Assoc. 2016;17(10):943-948.

25.

Adelborg

Horváth-Puhó

Ording

Pedersen

Sørensen

Henderson

. Heart failure and risk of dementia: A Danish nationwide population-based cohort study. Eur J Heart Fail. 2017;19(2):253-260.

26.

Manemann

Chamberlain

Bielinski

Jiang

Weston

Roger

. Predicting Alzheimer’s disease and related dementias in heart failure and atrial fibrillation. Am J Med. 2022;136(3):302-307.

27.

Jensen

Cenzer

Kelley

Covinsky

. Heart failure with comorbid dementia: Association with clinical characteristics and outcomes. J Am Geriatr Soc. 2023. Online ahead of print.

Supplementary Material

Please find the following supplemental material available below.

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

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

0.00 MB

0.35 MB