Measuring Cognition and Cognitive Impairment in the Survey of Health, Ageing and Retirement in Europe (SHARE): A Scoping Review and Instrument Mapping Study

Abstract

The Survey of Health, Ageing and Retirement in Europe (SHARE) is a cross-national panel study including approximately 160,000 adults aged ≥50 years from 29 countries. While multiple cognitive subtests are available, the SHARE consortium does not currently recommend a standardised approach to cognitive screening. This scoping review and mapping study aimed to (1) assess how cognition is measured in SHARE publications, (2) identify whether any cognitive screening instruments (CSIs) are validated in the SHARE, and (3) explore the potential to replicate additional CSIs using cognitive measures available in recent waves that include an expanded battery of subtests. SHARE-related publications were identified by searching PubMed, and a dedicated online registry of SHARE publications. Methodical details were extracted and quantitative counts calculated. Among 234 SHARE publications, the most common choices were using single subtests (n = 94), CSIs (n = 56), and standardised scores (n = 50). From 22 unique CSIs used in the SHARE, only the SHARE Cognitive Instrument and Langa–Weir Criteria were formally validated. Cognitive impairment was assessed in 36 studies, yet no validated recognised definition of mild cognitive impairment (MCI) was found. Mapping other potential CSIs (n = 81) identified the 10-Point Cognitive Screener, Six-Item Screener and Mini-Cog as other potential CSIs for use across SHARE waves. Further research is needed to validate existing CSIs and to better operationalise MCI in the SHARE.

1. Introduction

The prevalence of cognitive impairment among older adults is expected to rise globally in the coming decades [1,2] with particularly marked increases expected in European countries [3]. The development and validation of brief, reliable cognitive screening instruments (CSIs) for use within longitudinal ageing studies and in community-based samples is important for advancing our understanding of cognitive ageing at both the individual level and population level [4]. The use of short instruments can facilitate the study of its prevalence, risk factors, and progression across diverse populations [5]. They can also be used in clinical settings such as in primary care or general outpatient units to opportunistically screen individuals at risk for mild cognitive impairment (MCI), a prodromal stage before the onset of dementia, which identifies individuals who are at higher risk of developing dementia. Early intervention may support timely interventions to help mitigate the impact of cognitive decline on daily functioning and quality of life [6,7]. Brief CSIs (i.e., those taking ≤20 min to complete) can also be used as diagnostic tests as part of a more detailed comprehensive cognitive and functional assessment [5] and also to track decline in cognition over time.

The Survey of Health, Ageing and Retirement in Europe (SHARE) is a multi-country longitudinal study focused on middle-aged and older adults aged ≥ 50 years and includes questions assessing physical, psychological, cognitive and social factors [8]. Data collection began in 2004 and has (as of wave 9) spanned 29 countries including all continental European Union countries, Switzerland and Israel, which are included in the United Nations European Region. The SHARE uses standard questionnaires and in-person interviews based on the Health and Retirement Study (HRS) in the United States [9], which has inspired a network of sister studies globally [10,11]. The HRS network of studies has more recently started collecting detailed cognitive data through the Harmonized Cognitive Assessment Protocol (HCAP) project, acknowledging the need for more robust measures of cognitive impairment [12]. The first SHARE-HCAP study was conducted in five European Union countries in 2022 (data released 2025) [13]. While the SHARE consortium have made a probabilistic diagnosis of cognitive impairment within their HCAP study [13], no standardised measure of cognitive impairment has been recommended by the consortium for use in previous main (core) waves of the SHARE [14].

The core SHARE study includes a number of cognitive subtests across five cognitive domains including memory, language, orientation, visuospatial and executive function [14]. The number of subtests varies across waves, and the two most recent, waves 8 and 9, include 11 cognitive subtests such as delayed recall (10 word recall of a list of words) and verbal fluency (animal naming), amongst others. While these can be used alone or in combination to form proto instruments to assess cognition, no recognised standardised CSI has been recommended by the SHARE consortium to measure global cognitive function [14]. Thus, there is a lack of guidance on how cognition should be assessed within the SHARE. Efforts to support comparability across waves and between cohorts studies have focused predominantly on documenting the availability of individual cognitive subtests across studies [10,11,15,16], and to our knowledge, no study has examined which subtests or combination of subtests have been used in SHARE publications to measure cognition.

The SHARE, unlike some HRS-based surveys of ageing [11,14], did not originally include any complete established CSIs for global cognition such as the Mini Mental State Examination (MMSE) [17] or the Montreal Cognitive Assessment (MoCA) [18], the Telephone Interview for Cognitive Status (TICS) [19] or the Community Screening Instrument for Dementia (CSI-D) [20]. However, given the plethora of short CSIs that have been proposed in the literature [21,22], there may be multiple brief CSIs that could be replicated in the SHARE using the subtests provided. To identify these, and how researchers have tackled this deficit to date, a systematic overview of the existing literature is required. Scoping reviews are designed to summarise broad topics or fields in order to clarify working definitions, identify research gaps and make recommendations for future research [23]. Hence, we conducted a scoping review and instrument mapping study to address three objectives aimed at providing more clarity and consistency regarding the measurement of cognition in the main waves of the SHARE:

(1)

To review how cognitive function and cognitive impairment have been assessed in previous SHARE publications including any details on what subtests were used, any cut-offs applied, the number of studies using the approach, and the longitudinal availability of each approach across study waves.

(2)

To identify which, if any, CSIs and measures of cognitive impairment have been validated internally or externally for use in the SHARE.

(3)

To explore which additional CSIs could be operationalised for use within the SHARE based on cognitive subtests (items) available and cross-referenced with previously published reviews of brief CSIs.

2. Materials and Methods

This study involved three steps: (1) a scoping review of published SHARE studies examining cognition; (2) a review of brief CSIs identified from the broader literature; and (3) an instrument mapping exercise to assess which CSIs could be operationalised using available SHARE subtests. This mixed-methods approach is outlined in Figure 1.

In the first step, the SHARE register and PubMed were searched to identify SHARE studies that assessed cognition and were published between 2003 and the 25 July 2025. The SHARE online register (available at: https://share-eric.eu/publications accessed on 25 July 2025) documents studies that use SHARE data. PubMed was added as a second source after identifying that some studies were not present in the register at this time. Separate to this, in the second step, short (less than 30 min to administer) CSIs from the wider literature were identified by a search of existing narrative and systematic review papers examining brief CSIs. In the third step, an instrument mapping exercise was conducted, which reviewed all the CSIs identified in the previous step and what items (i.e., cognitive subtests) and which cognitive domains these included. An assessment of their fidelity (indicating their completeness with the original instrument compared to what is available in the SHARE) against the subtests available in the SHARE was carried out. The scoping review component of this study was conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) [24] (checklist provided in the appendix). A protocol was published on Open Science Framework (https://doi.org/10.17605/OSF.IO/NTPMF registered on 21 May 2025). A PRISMA flow diagram was produced to summarise the study selection and exclusion using the R package “PRISMA2020” [25,26].

2.1. Eligibility Criteria for the Scoping Review

This review included studies assessing cognition in the SHARE published using SHARE data since inception (2003). The SHARE studies were limited to primary research papers assessing cognitive impairment or function (cognitive) using secondary data from the SHARE. It could include any primary study which proposed a unique short CSI for use in this cohort. Only studies that used an objective measure of cognition were included; those using only subjective or informant rated measures of cognition were excluded such as self-rated memory [27] or the Informant Questionnaire on Cognitive Decline in the Elderly [28]. Studies from any year and published in any language were considered. Publications in languages other than English were translated using Google Translate. Included studies were limited to full peer-reviewed articles. We excluded abstract publications, book chapters, protocols, reviews and SHARE working documents for several reasons including concerns over methodological rigour and a focus on describing the SHARE study itself. A list of all the excluded studies is provided in the Supplementary Material Table S1.

2.2. Information Sources, Search Terms and Screening for the Scoping Review

The SHARE studies were identified in three ways: from an online register of SHARE studies, from a search of the PubMed database and from citation tracking of included studies. The search was updated on 25 July 2025. The SHARE study register (https://share-eric.eu/publications accessed on 25 July 2025) contains over 4000 submissions, and potential papers were identified by applying the following search terms without any filters: “cognitive”, “cognition”, “cognitive impairment”, “cognitive screening”, “cognitive assessment” and “cognitive screening instruments”. This register should include a complete list of SHARE studies, given it is required that all SHARE studies are reported to the SHARE consortium in the conditions for use [29]. To supplement this, PubMed was also searched using the following string: “Survey of Health Ageing and Retirement in Europe” AND (“cognitive” OR “cognition” OR “cognitive impairment” OR “cognitive screening” OR “cognitive assessment” OR “cognitive screening instruments”). Study citations were copied to an Excel spreadsheet and were screened by a single reviewer. Most ineligible study types (e.g., book chapters and working documents) could be identified from details in the SHARE database and could thus be excluded at the title screening stage. The SHARE database provided the study title, a citation/reference and the publication year. The rest of the studies were moved to full-text review.

2.3. Data Extraction (Charting) Process and Items

A data-charting form was jointly developed by two reviewers (MO’D and RO’C) to determine which variables to extract as part of the scoping review. This included descriptive details (study identifier, study title, a citation/reference, and publication year), screening details (study design), and cognitive methods (subtests used, CSI used, another details on methods and cut-offs scores). One reviewer charted the data, and both reviewers discussed the results and continuously updated the data-charting form in an iterative process. Where necessary, methodological details were extracted from the articles’ Supplementary Material or citated papers. The range of scores could also be used to differentiate SHARE subtests including serial 7s (scored 0–5) and numeracy (scored 1–5).

2.4. Instrument Mapping Study

The instrument mapping exercise was based on a search of two large reviews of short CSIs including a 2007 narrative review describing the psychometric characteristics of 39 CSIs [21] and a more recent systematic review published in 2019, which included 50 CSIs [22]. These were identified based on a limited search of Google Scholar using the term “brief cognitive screening instrument”. Hand searching included any additional CSIs known to the study authors that were not listed in either of these review papers. Details of the instruments including their subtests and scoring structure were extracted. Instruments were excluded if they were informant-generated, had an administration time more than 20 min long or included only a single subtest. For CSIs identified in the literature, their fidelity within the SHARE was assessed as the proportion of the total score which could be replicated using items available in the SHARE (including item substitution if very similar cognitive subtests were available).

2.5. Synthesis of Results

This study was designed to identify cognitive screening approaches and CSIs that are available or could be developed for use in the SHARE; hence, a quality appraisal of the included studies was not considered. For the quantitative analysis, the number and proportion of studies using each approach were calculated from the extracted details using R version 4.4.2 [30].

2.6. Terminology and Available Subtests

General cognitive performance was divided into five domains to reflect those available in the SHARE, i.e., memory, language/fluency, orientation, executive functioning and visuospatial, which is consistent with HCAP studies, a cognitive sub-study replicated across HRS-related studies [31,32,33]. Memory was further subdivided into immediate and delayed memory [31]. There are 11 cognitive subtests available in the SHARE across waves 1–9 derived from different cognitive tests including the MMSE [17], TICS [19], (CSI-D) [20] and Addenbrooke’s Cognitive Examination III (ACE-III) [34]. These include the following cognitive subtests: 10-word immediate recall [registration], 10-word (delayed) recall, animal naming, time orientation (date, day, month, year), serial 7s, numeracy (4 items), counting backwards, copying cube, copying infinity loop, clock drawing, and object naming (3 items). These 11 items are not available across all waves of the SHARE with an increasing number added over waves. For this review, the term “word registration” is used for the 10-word immediate recall test, which some studies have also termed verbal learning. These subtests and their availability across waves of the SHARE are summarised in

Table 1. Overview of subtests available in the Survey of Health, Ageing and Retirement in Europe (SHARE) including which waves (w) they are available in.

Alternative Names/Details	Subtest Description	Original Instrument	w1	w2	w3	w4	w5	w6	w7	w8	w9
10-word registration (immediate recall or verbal learning)	The participant is read a list of 10 words and asked to recall as many as possible. One attempt is given, but the need to remember the words is explained before starting. One of 4 random word lists is selected.	TICS	✓	✓	X	✓	✓	✓	✓	✓	✓
10-word recall (delayed recall)	The participant is asked to recall as many of the words as they can from the 10-word registration task earlier.	TICS	✓	✓	X	✓	✓	✓	✓	✓	✓
Verbal fluency (semantic)	Animal naming in exactly one minute (timed). All creatures within the animal kingdom, mythical, species, breeds, as well as male, female, infant name variations were accepted (max 100). Excluded repetitions and proper nouns.	Henley et al. 1969 [35]	✓	✓	X	✓	✓	✓	*	✓	✓
Orientation (4 items)	Four temporal orientation questions were asked: date, month, year, day of the week.	MMSE	✓	✓	X	*	*	✓	*	✓	✓
Serial 7s (calculation)	Numerical subtraction question: “One hundred minus 7 equals what? And 7 from that, And 7 from that, And 7 from that, And 7 from that.” Scored from 0 to 5.	TICS MMSE	X	X	X	✓	✓	✓	*	✓	✓
Numeracy (5 items)	Numerical percentage questions scored from 1 to 5 [note the score may also be recoded as 0 to 4 in SHARE publications].	Unique to SHARE	✓	✓	X	*	*	*	*	*	*
Counting backwards	The participant is asked to count backward as quickly as possible starting at 20. Correct if counted from 19 to 10 or from 20 to 11 without error. Option for a second attempt was given.	TICS	X	X	X	X	X	X	X	60+	60+
Copying infinity loop	The participant is asked to draw a copy of an infinity loop diagramprovided to them. A second attempt was allowed.	ACE-III	X	X	X	X	X	X	X	*60+	*60+
Copying cube	The participant is asked to draw a copy of a cube diagram provided to them. A second attempt was allowed.	ACE-III	X	X	X	X	X	X	X	*60+	*60+
Clock drawing test	The participant is asked to draw a clock face (second attempt allowed). If the contour or numbers were drawn correctly then the patient was asked to add hands pointing to ten past five.	ACE-III	X	X	X	X	X	X	X	*60+	*60+
Object naming (3 items)	This is a language rather than a visual test. The three objects to be named were as follows: scissors: “What do people usually use to cut paper?”, cactus: “What do you call the kind of prickly plant that grows in the desert?”, pharmacy: “Where do people usually go to buy medicine?”. Synonyms and specific cactus names were accepted.	TICS (scissors, cactus) and CSI-D (pharmacy)	X	X	X	X	X	X	X	*65+	*65+

✓ denotes that an item was assessed in that wave. * Denotes that an item was assessed for a subsample of SHARE participants in that wave. X denotes that an item was not assessed at that wave. For example, wave 3 (SHARELIFE) did not include any cognitive assessments as it focused on childhood conditions; and wave 7 includes a combination of SHARELIFE and the main questionnaire and had fewer cognitive assessments. Some items were limited to those aged ≥60 years or ≥65 years. ACE-III: Addenbrooke’s Cognitive Examination III [34]; CSI-D: Community Screening Instrument for Dementia [20]; MMSE: Mini-Mental State Examination [17]; TICS: Telephone Interview for Cognitive Status [19].

.

3. Results

From a total of 442 reports found in the search of the SHARE register and PubMed (step 1), there were 234 studies eligible for inclusion. Thirty-four full texts were excluded at screening for the following reasons: did not assess cognition (n = 25), did not use SHARE data (n = 4), abstract only publication (n = 2), a review article (n = 2), and a trial protocol (n = 1). Four of the included studies were published in languages other than English including Hebrew (n = 1), Bulgarian (n = 1), German (n = 1), and Spanish (n = 1) and were translated using Google Translate. In the separate search (step 2), after removing duplicates (n = 10), informant-reported measures (n = 7) and screens based on single subtest items (n = 11), a total of 61 unique CSIs were identified [21,22]. The selection of studies is detailed in two PRISMA (2020) flow diagrams (Figure 2). Data extracted for the included studies is provided in the Supplementary Material Tables S2 and S3. Of the 234 identified studies 146 (62%) came from the SHARE register, 77 (33%) from PubMed and 11 (5%) from citation tracking.

3.1. Approaches to Measuring Cognition in the SHARE Studies

Among the 234 SHARE studies, 140 (60%) applied one or more composite scores (i.e., multiple subtests combined) and 94 (40%) provided results for individual cognitive subtests only. A full breakdown of the methods utilised in the SHARE studies to assess cognition is provided in Figure 3.

3.2. Cognitive Subtests Used in the SHARE

The subtests most commonly applied in the SHARE studies are displayed below (

Table 2. Overview of the cognitive domains and subtests applied in published SHARE studies.

Cognitive Domain Cognitive Test	All (N = 234)	Subtest Only (n = 94)	Composite Scores (n = 140)	CSIs (n = 56) *	Standardised Scores (n = 50) *	Statistical Modelling (n = 17) *	Cut-Offs (n = 15)	Other Methods (n = 5)
Memory	223 (95%)	85 (90%)	138 (99%)	56 (100%)	49 (98%)	17 (100%)	14 (93%)	5 (100%)
Registration (immediate word recall)	193 (82%)	56 (60%)	137 (98%)	56 (100%)	48 (96%)	17 (100%)	14 (93%)	5 (100%)
Delayed word recall	217 (93%)	79 (84%)	138 (99%)	56 (100%)	49 (98%)	17 (100%)	14 (93%)	5 (100%)
Language/fluency	180 (77%)	68 (72%)	112 (80%)	33 (59%)	50 (100%)	14 (82%)	13 (87%)	4 (80%)
Verbal fluency	180 (77%)	68 (72%)	112 (80%)	33 (59%)	50 (100%)	14 (82%)	13 (87%)	4 (80%)
Naming descriptions	1 (<1%)	-	1 (1%)	1 (2%)	-	-	-	-
Orientation	60 (26%)	14 (15%)	46 (33%)	15 (27%)	18 (36%)	2 (12%)	7 (47%)	4 (80%)
Date/day	60 (26%)	14 (15%)	46 (33%)	15 (27%)	18 (36%)	2 (12%)	7 (47%)	4 (80%)
Executive functioning	99 (42%)	28 (30%)	71 (51%)	32 (57%)	23 (46%)	10 (59%)	4 (27%)	4 (80%)
Clock drawing	1 (<1%)	-	1 (1%)	1 (2%)	-	-	-	-
Serial 7	53 (23%)	11 (12%)	42 (30%)	23 (41%)	11 (22%)	4 (24%)	2 (13%)	4 (80%)
Numeracy	51 (22%)	18 (19%)	33 (24%)	10 (18%)	12 (24%)	8 (47%)	2 (13%)	1 (20%)
Counting backwards	2 (1%)		2 (1%)	2 (4%)	-	1 (6%)	-	-
Visuospatial	1 (<1%)		1 (1%)	1 (2%)	-	-	-	-
Copy cube	1 (<1%)		1 (1%)	1 (2%)	-	-	-	-
Copy infinity loop	1 (<1%)		1 (1%)	1 (2%)	-	-	-	-

* Three studies were included in more than one category including two studies with CSIs and statistical modelling [33,37] and another one included in both CSIs and standardised scores [36]. Note this table includes all subtests used in a study not necessarily how they were used, for example, one study includes both the SHARE-Cog and a non-amnestic CSI [38] so the subtests from both are included in the count under CSIs. Bold text denotes a cognitive domain and the regular text describes the subtests within that domain.

) and are divided according to each of the categories in Figure 3. Only one study [38] included any of the newly available subtests from the latter waves of the SHARE such as naming descriptions, the clock drawing test, copying cube or the infinity loop, as these were only introduced at wave 8 (data released on 10 February 2022). Reflecting this, the most commonly used subtests overall were word registration, word recall and verbal fluency (each found in over 75% of studies). Other measures (orientation, serial 7s or numeracy tests) were each used in 26% of the studies or less. As noted, only one study utilised the recent subtests assessing visuospatial function.

3.3. Use of Individual Cognitive Subtests (n = 94)

A total of 94 studies presented results for individual cognitive subtests (items). Of these, only 8 (9%) used cut-offs to define a deficit on one or more of the included subtests. These cut-offs were obtained from either previous works (n = 5; three SHARE reports/papers, one non-SHARE studies, one uncited) or from pre-determined proportions (n = 3) taking either the lowest half (based on the median), the lowest one third (or as close to this share as possible) or the 25th percentile. The number of subtests included ranged between one (19 studies, 20%) and five (9 studies, 10%). Amongst the 19 studies which only included a single subtest, the majority chose to include word recall (n = 10) followed by word registration (n = 4), verbal fluency (n = 3) or orientation (n = 2). One study used word registration scores to compare memory across both the SHARE and the China Health and Retirement Longitudinal Study [39]. Some studies (n = 6), which included multiple subtests, decided to present standardised z-scores instead of the raw scores to make easier comparisons between the subtest scores. Only one study included cognitive subtests as independent variables within a statistical modelling (Random Forest algorithm) [40].

3.4. Use of Cognitive Screening Instruments (n = 56)

In total, 56 SHARE studies included at least one CSI derived from multiple subtests. Most (n = 30) focused on memory, combining 10-word registration and delayed recall into a 20-point sum score (n = 27) or a 10-point mean score (n = 3). Only two of these studies applied cut-offs to define memory impairment: one used a threshold of 1.5 SDs below age-specific means [41], while another incorporated the recall score into an algorithm for probable dementia that also included activities of daily living and country-level dementia prevalence estimates [37]. The remaining studies (n = 26) combined subtests across domains to create 20 distinct CSIs (

Table 3. Overview of cognitive screening instruments for the SHARE including their availability and cognitive domains assessed.

Cognitive Screening Instruments (CSIs) in SHARE Studies (n = 22)	Number of Studies (n = 56)	Number of Waves Fully Available	Wave 1	Wave 2	Wave 4	Wave 5	Wave 6	Wave 7	Wave 8	Wave 9	Number Cognitive Domains	Memory	Language	Orientation	Executive Function	Visuospatial
Validated internally/externally (n = 2)
SHARE-Cog (0–45) [38]	1 **	7	✓	✓	✓	✓	✓	*	✓	✓	2	✓	✓	X	X	X
Langa–Weir criteria (0–27) [33]	1	0	X	X	X	X	X	X	*	*	2	✓	X	X	✓	X
Used in multiple studies (n = 5)
CSI 0–125 (serial 7s) [44,45,46,47]	4	5	X	X	✓	✓	✓	*	✓	✓	3	✓	✓	X	✓	X
CSI 0–129 [48,49]	2	2	✓	✓	*	*	*	*	*	*	4	✓	✓	✓	✓	X
CSI 0–25 [50,51]	2	5	X	X	✓	✓	✓	*	✓	✓	2	✓	X	X	✓	X
CSI 0–29 (serial 7s) [52,53]	2	3	X	X	*	*	✓	*	✓	✓	3	✓	X	✓	✓	X
DemTect modified (0–20) [54,55]	2	2	✓	✓	*	*	*	*	*	*	4	✓	✓	✓	✓	X
Used in a single study (n = 13)
Recall and fluency (0–65) [58]	1	7	✓	✓	✓	✓	✓	*	✓	✓	2	✓	✓	X	X	X
Recall and fluency (0–20) [59]	1	7	✓	✓	✓	✓	✓	*	✓	✓	2	✓	✓	X	X	X
Recall and fluency (0–20 weighted) [60]	1	7	✓	✓	✓	✓	✓	*	✓	✓	2	✓	✓	X	X	X
Recall and fluency (0–30 weighted) [36]	1	7	✓	✓	✓	✓	✓	*	✓	✓	2	✓	✓	X	X	X
CSI 0–35 [61]	1	5	X	X	✓	✓	✓	*	✓	✓	3	✓	✓	X	✓	X
Langa–Weir modified (0–26) [57]	1	5	X	X	✓	✓	✓	*	✓	✓	3	✓	✓	X	✓	X
CSI 0–39 [62]	1	3	X	X	*	*	✓	*	✓	✓	4	✓	✓	✓	✓	X
CSI 0–34 [63]	1	5	✓	✓	*	*	✓	*	✓	✓	3	✓	✓	✓	X	X
CSI 0–70 [64]	1	5	X	X	✓	✓	✓	*	✓	✓	3	✓	✓	X	✓	X
CSI 0–29 (numeracy) [65]	1	2	✓	✓	*	*	*	*	*	*	3	✓	X	✓	✓	X
CSI 1–125 (numeracy) [66]	1	2	✓	✓	*	*	*	*	*	*	3	✓	✓	X	✓	X
DemTect modified (0–19) [56]	1	2	✓	✓	*	*	*	*	*	*	4	✓	✓	✓	✓	X
Non-amnestic battery (0–16) [38]	1 **	2	X	X	X	X	X	X	✓	✓	4	X	✓	✓	✓	✓
Monodomain—2 tests (n = 2)
20-point word recall test	27	8	✓	✓	✓	✓	✓	✓	✓	✓	1	✓	X	X	X	X
10-point average recall [67,68,69]	3	8	✓	✓	✓	✓	✓	✓	✓	✓	1	✓	X	X	X	X

X not available. * available for a subset of participants. ✓ available for all. ** One study had two CSIs [38]. Bold text describes key features of the CSIs relating to their validity.

), of which only two were formally validated: the Langa–Weir criteria [42] and the SHARE Cognitive Instrument (SHARE-Cog) [38]. The Langa–Weir criteria includes four subtests (10-word registration/recall, serial 7s and counting backwards) [42] and has been predominantly used in the HRS in the United States and used data from the Aging, Demographics, and Memory Study (ADAMS) sub-study, which included gold-standard clinical diagnoses [42]. However, these subtests are only available in the SHARE for participants aged ≥ 60 years in waves 8–9 and have not been validated in European samples. The SHARE-Cog combines registration, recall and verbal fluency, is available across all SHARE waves, and was validated internally in participants aged ≥ 65 (using wave 8), showing excellent accuracy for cognitive impairment and good discrimination between MCI and subjective memory complaints [38]. Cognitive impairment was defined using additional wave-8 subtests and self-reported activities of daily living (ADL) and memory difficulties consistent with Petersen’s criteria [38,43]. To avoid incorporation bias with the items included in the SHARE-Cog, the definition of MCI was restricted to the remaining (non-amnestic) subtests available in wave 8 of the SHARE. Five CSIs were used in more than one study, including a 125-point CSI [44,45,46,47], a 129-point CSI [48,49], a 25-point CSI [50,51], a 29-point CSI [52,53], and a modified 20-point DemTect [54,55]. The 125-point CSI was most common, summing registration, recall, verbal fluency and serial 7s [44,45,46,47]. Modified DemTect versions used in early waves [54,55,56] and a modified 26-point Langa–Weir including verbal fluency [57] had limited applicability, while all remaining CSIs (n = 11) were used in single studies only. None of the identified CSIs included all 5 cognitive domains.

3.5. Identification of Additional CSIs for the SHARE (n = 24)

Of the 61 other CSIs identified from the literature, there were 24 that could be replicated using subtests available in the SHARE to at least 50% fidelity with the original. Of these, three had 100% fidelity (no missing items but subtests could be substituted with very similar items if required), namely the 10-point Cognitive Screener [70], Mini-Cog [71], and Six Item Screener [72], although the 3-word recall subtest had to be replaced with the 10-word recall test available in the SHARE. Potentially available short CSIs that have not been included in the SHARE to date are presented in

Table 4. Overview of additional potential cognitive screening instruments with at least 50% fidelity in the SHARE including their longitudinal availability and the subtest domains assessed in the SHARE (n = 24).

Cognitive Screening Instrument (Potentially Available But Not Used in SHARE Yet)	Fidelity 1 in the SHARE	Number of Waves Fully Available	Wave 1	Wave 2	Wave 4	Wave 5	Wave 6	Wave 7	Wave 8	Wave 9	Number Cognitive Domains	Memory (SHARE)	Language (SHARE)	Orientation (SHARE)	Executive function (SHARE)	Visuospatial (SHARE)
10-point Cognitive Screener (10-CS) [70]	100%	5	✓	✓	*	*	✓	*	✓	✓	3	✓	✓	✓	X	X
Mini-Cog [71]	100%	0	X	X	X	X	X	X	*	*	2	✓	X	X	✓	X
Six Item Screener (SIS) [72]	100%	5	✓	✓	*	*	✓	*	✓	✓	2	✓	X	✓	X	X
AB Cognitive Screen (ABCS) 135 [74]	96%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Rapid Cognitive Screen [75]	90%	0	X	X	X	X	X	X	*	*	2	✓	X	X	✓	X
Computer test battery by Inoue et al. [76]	87%	5	✓	✓	*	*	✓	*	✓	✓	2	✓	X	✓	X	X
Computer Self-Test [77]	83%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Short MoCA [78]	81%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Montpellier Screen [79]	80%	0	X	X	X	X	X	X	*	*	2	X	✓	X	X	✓
DemTect [80]	75%	7	✓	✓	✓	✓	✓	*	✓	✓	3	✓	✓	X	✓	X
Brief Alzheimer Screen [81]	73%	5	✓	✓	*	*	✓	*	✓	✓	3	✓	✓	✓	X	X
Test for the Early Detection of Dementia from Depression [82]	73%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Concise Cognitive Test [83]	70%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	X	✓
Montreal Cognitive Assessment (MoCA) [18]	67%	0	X	X	X	X	X	X	*	*	5	✓	✓	✓	✓	✓
MoCA computer tool [84]	67%	0	X	X	X	X	X	X	*	*	5	✓	✓	✓	✓	✓
Quick Mild Cognitive Impairment (Qmci) Screen [85]	67%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Rapid Dementia Screening Test [86] 2	67%	7	✓	✓	✓	✓	✓	*	✓	✓	1	X	✓	X	X	X
Rowland Universal Dementia Assessment Scale (RUDAS) [87]	63%	0	X	X	X	X	X	X	*	*	2	✓	✓	X	X	✓
Mini-Mental State Examination (MMSE) [17]	60%	0	X	X	X	X	X	X	*	*	5	✓	✓	✓	✓	✓
Mini-Revised Kingston Standardized Cognitive Assessment [88]	55%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Telephone Interview of Cognitive Status (TICS)—Modified [89]	54%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
M-ACE [90]	53%	0	X	X	X	X	X	X	*	*	3	X	✓	✓	✓	X
MoCA Basic Version [91]	53%	0	X	X	X	X	X	X	*	*	4	✓	✓	✓	✓	X
Short and Sweet Screening Instrument [73]	53%	0	X	X	X	X	X	X	*	*	5	✓	✓	✓	✓	✓

¹ Represents how much of the total score could be replicated. ² Only verbal fluency was available. X not available. * available for a subset of participants. ✓ available for all.

along with the waves they could be generated in, acknowledging that most (n = 18/24) were only partially available in waves 8 and 9 and were not available in any other waves of the SHARE, limiting their utility to measure cognition over time. CSIs based on the MMSE [17,73] or MoCA [18] would include all five cognitive domains available in the SHARE.

3.6. Use of Standardised Cognitive Scores (n = 50)

In total, 50 studies created global cognitive scores by statistically standardising subtests. Global cognition (50 studies) was measured by statistically standardising each subtest and then combining them into a single composite score. The most common standardization approach was to take the average of multiple z-scores (n = 38); other approaches included taking the sum of multiple z-scores (n = 5) or using them to calculate a T-score (n = 7). A z-score is a standardized score that has a mean of 0 and a standard deviation (SD) of 1 in the reference group; and is calculated by subtracting the reference mean from each raw value and then dividing by the reference SD [92]. For the SHARE studies, all composite z-scores were composed of between 3 and 5 subtests. T-scores are generated using z-scores and have a mean of 50 and a SD of 10 [93]. Seven SHARE studies generated a composite T-score to measure cognition including a sum of the z-scores of 3–4 subtests which were standardised according to the mean and SD of those aged 50–54 years. T-scores have also been applied in the recent SHARE-HCAP study to define MCI and probable dementia [13,33].

3.7. Use of Statistical Modelling (n = 17)

There were 17 studies that used statistical modelling (including regression-based approaches, machine learning algorithms, and dimensionality-reduction techniques) to measure global cognition. Most of these used either principal component analysis (n = 9) or an unsupervised machine learning approach based on principal components (n = 3). Principal component analysis is a linear dimensionality-reduction technique that maximizes the variance of the data projected onto the principal components [94]. In other words, it finds a smaller number of new variables (called “components”) that summarize the original data while retaining as much information (i.e., variance) as possible. Factor analysis was also applied including exploratory factor analysis [95] and longitudinal modelling approaches such as a cross-lagged model [96] and a second-order latent growth curve model [97]. Another study fitted a regression model using the HCAP classification to identify the probability of an individual having MCI or dementia using a combination of cognitive subtest scores and the presence (i.e., dementia) or absence (i.e., MCI) of ADL impairment [33], and found that the model had good validity compared with HCAP prevalence values. Finally, a study exploring methods for identifying probable dementia by Klee et al. [37] applied a number of different models and utilised the word recall components of the Langa–Weir (i.e., 20-point word recall), with the number of ADL limitations to define probable dementia. Their relatively simple “Langa-Weir” model adjusted to Organisation for Economic Co-operation and Development (OECD) data outperformed other, more sophisticated methods, including random forest modelling [37]. Random forest models are machine learning algorithms which combine values from multiple randomised decision trees into a single result [98].

3.8. Using Cut-Offs to Measure Global Cognitive Impairment (n = 15)

There were 15 studies that defined cognitive impairment as the presence of impairments in multiple subtests (i.e., cognitive items) using cut-offs. Fixed threshold scores were used in two studies [99,100]. Another used cut-offs based on the number of impaired subtests (e.g., ≥2 subtests below threshold [101] or three impaired subtests [102]). The multi-country ATHLOS Project used proportional cut-offs with the lowest 25% of scores considered to represent cognitive impairment [16]. One study took account of the absolute number of impaired subtests from 0 to 4 [103]. The remaining nine studies all took cut-offs of 1.5 SDs below a mean score.

3.9. Other Approaches (n = 5)

Other methods used to measure cognition were applied in five studies including averaging subtest scores (n = 1) [104], the use of deficit accumulation indices (n = 3) [105,106,107], and meta-analysis (n = 1) [108]. For the study that used average raw scores, a sum of registration, recall, and verbal fluency was taken and divided by three [104]. The deficit accumulation approach [105,106,107], which is based on a frailty index, scores the level of risk using the proportion of deficits present [109]. This is an unusual approach in the context of cognitive assessment but could be considered to be measuring cognitive frailty as a construct. The meta-analysis approach combined Cohen’s d scores from six SHARE subtests (registration; recall, verbal fluency, orientation, serial 7s and numeracy) [108].

3.10. Use of Cut-Offs

The most frequently applied cut-offs for specific cognitive subtests were <5 words for registration (8/19 studies), <4 words for recall (9/18 studies), and <15 for verbal fluency (8/22 studies). Fewer studies were available for other subtests with the most frequent cut-offs being <4/4 (5/9 studies) for orientation, <4/5 (3/6 studies) for serial 7s and <2/5 (2/4 studies) for numeracy. While verbal fluency is scored from 0 to 100 in the SHARE, some studies rescaled it from 0 to 10 (divided by 10) or applied upper limits such as ≤45 words (n = 16 studies). Other approaches to cut-offs included taking scores <1.5 SDs below the mean, 25th percentile and the lowest 7% of scores (full results in the Supplementary Material Table S4).

3.11. Classifying Stages of Cognitive Impairment

The majority of studies only presented mean CSI scores representing cognitive functioning. As illustrated in Table 5, only 36 (26%) of the 140 composite measures of cognition divided participants into different cognitive groups (i.e., stages) based on their level of cognitive impairment, including a total of 46 different measures: 21 for probable dementia, 2 for cognitive impairment no dementia (CIND), 4 for MCI, 17 for CI (MCI or dementia) and 2 for amnestic impairment. For probable dementia, one paper compared 14 approaches including statistical models and CSI cut-offs with/without ADLs [37]. They recommended using a “Langa–Weir algorithm” which applied OECD-adjusted cut-offs to 20-word recall and country-specific cut-offs for the number of instrumental ADL (IADL) difficulties [37], which produced favourable results compared with the more sophisticated models assessed. Two other studies also used a cognitive cut-off score together with the presence of difficulties completing ADLs to identify dementia [38,110], which is consistent with the DSM-4/DSM-5 criteria for dementia and major neurocognitive disorders [111]. Another approach applied was a multivariate regression model fitted to HCAP data which was then applied to calculate probabilities of MCI and dementia for each participant in an earlier wave of the SHARE [33]. Finally, four studies [33,51,57,112] defined dementia using only cognitive cut-off scores. One of these, the Langa–Weir criteria, used a cut-off which has been validated in the Aging, Demographics, and Memory Study (ADAMS) in the United States [42], and another used a modification of this with the same cut-offs [57]. Two studies identified dementia using a cut-off 1.5 SD below a mean score [51,112], which is often considered more reflective of mild impairment [111].

The Langa–Weir criteria also applies a higher cut-off range to identify those with CIND, an approach that has also been validated with ADAMS data [42]. A second study applied these same cut-offs on a similar CSI [57]. For MCI there was one study which approximated Petersen’s criteria [43], taking a cut-off for cognitive impairment on a defined cognitive battery of subtests, the absence of ADL difficulties, presence of self-reported memory complaints and the absence of dementia [38]. A second study used the HCAP sample and a statistical model to calculate the probability of having MCI for each participant [33]. Four other studies [113,114,115,116] applied a cut-off of 1.5 SDs below a mean to identify MCI, without excluding dementia or examining ADLs. Almost half (11 of 23) of studies defining CI (MCI or dementia) used cut-offs based on SDs below mean scores with adjustments for key variables such as age and education. However, none of these studies used a normative sample which may lead to erroneous results.

Table 5. Definitions of cognitive categories and stages identified in the SHARE studies classifying participants with probable dementia, cognitive impairment no dementia (CIND), mild cognitive impairment (MCI), and cognitive impairment (CI) defined as MCI or dementia.

Probable Dementia Criteria Description	Studies (n = 7) Criteria (n = 21)	Cognitive Measure Classification Approach	Terminology Used	Validation Status Notes
Langa–Weir criteria [33] Langa–Weir (registration, recall, serial 7s, counting backwards) ≤ 6/27	1	CSI cut-off Selected cut-off	Probable dementia	Validated in United States Best cut-off based on a gold standard diagnosis in the ADAMS [33].
Langa–Weir criteria (modified) [57] CSI (registration, recall, serial 7s, verbal fluency) ≤ 6/26	1	CSI cut-off Selected cut-off	CI consistent with dementia	Not validated Cut-off from Langa–Weir. (swapped 2-point counting with 1-point fluency).
Modified DSM-5 [38] Non-amnestic battery ≥ 2 SDs below mean by age/education (or self-reported dementia/memory disorder) with self-reported IADL difficulty (≥1 of telephone calls, taking medications, managing finances)	1	CSI cut-off SDs below mean + ADLs	Dementia	Not validated Cognitive battery did not include memory issues. IADL difficulty for cognitively orientated tasks only, not necessarily loss of independence. Prevalence was low.
LW algorithms (X2) [37] 20-point word recall 2.5% percentile (i.e., ≥2 SD below mean). Applied with or without IADL difficulties taking a cut-off of 1.5 interquartile ranges above Q3 (by country)	1	CSI cut-off Percentile + ADLs	Probable dementia	Percentile LW algorithms with ADLs recommended Cognitive battery only included memory issues. IADL difficulty had different cut-offs by country, not loss of independence. Performed well in this study.
1.5 SD below adjusted mean [51] 1.5 SD below 25-point CSI (registration, recall, serial7s) by age, education, sex, proxy status of interviews	1	CSI cut-off SDs below mean	Probable dementia	Not validated No normative sample used for memory. 1.5 SDs below a mean for MCI/CIND.
LW algorithms (X2) [37] Used 20-point word recall cut-offs from equipercentile with country prevalence values in OECD. Applied with or without IADL difficulties taking a cut-off of 1.5 interquartile ranges above Q3 (by country)	1	CSI cut-off equipercentile	Probable dementia	Percentile LW algorithms with ADLs recommended Cognitive battery only included memory issues. IADL difficulty had different cut-offs by country, not loss of independence.
Cut-off/1.5 SDs below adjusted mean [112] Memory (20-word recall) 1.5 SDs below the mean by age or verbal fluency < 15	1	Cut-offs (domains) SDs below mean / cut-off	Dementia	Not validated No normative sample used for memory. Not adjusted for age/education for fluency. 1.5 SDs below a mean for MCI/CIND.
1.5 SDs below mean [110] Standardised score (registration, recall, fluency, orientation) 1.5 SDs below mean and BADL difficulty (bathing, eating, dressing, transferring, or walking) or self-reported dementia	1	Standardised score SDs below mean + ADLs	Dementia	Not validated Not adjusted for age/education.
Random forest (X3) [37] Random Forest fitted using three training dataset approaches (registration, recall, ADLs)	1	Statistical model Machine learning	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
XGBoost classifier (X3) [37] XGBoost classifier in three training dataset approaches (registration, recall, ADLs)	1	Statistical model Machine learning	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Logistic regression model (X3) [37] logistic regression fitted to self-reported cases (registration, recall, ADLs)	1	Statistical model Regression	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Weighted logistic regression [37] Weighted logistic regression fitted to self-reported cases (registration, recall, ADLs)	1	Statistical model Regression	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Predicted probabilities [33] (SHARE-HCAP) A predictive model was fitted between wave 9 items (cognition, ADLs, etc.) and HCAP cognitive categories	1	Statistical model Multivariate regression	Dementia	Categorical status not validated Average of predicted probabilities matched HCAP prevalences very well. Strong association with education.
CIND criteria Description	Studies (n = 2) Criteria (n = 2)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Langa–Weir criteria [33] Langa–Weir (registration, recall, serial 7s, counting backwards) 7–11 out of 27	1	CSI cut-off Selected cut-off	CIND	Validated in United States Best cut-off based on a gold standard diagnosis in the ADAMS [33].
Langa–Weir criteria (modified) [57] CSI (registration, recall, serial 7s, verbal fluency) 7–11 out of 26	1	CSI cut-off Selected cut-off	Probable dementia	Not validated Cut-off from Langa–Weir. (swapped 2-point counting with 1-point fluency).
MCI criteria Description	Studies (n = 6) Criteria (n = 4)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Modified Petersen’s criteria [38] Non-amnestic battery and self-reported questions on ADLs, memory, dementia used to define subjective memory complaints, MCI and dementia	1	CSI cut-off SDs below mean	MCI	Not validated MCI definition incomplete—missing amnestic MCI single domain.
Predicted probabilities [33] (SHARE-HCAP) A predictive model was fitted between wave 9 items (cognition, ADLs, etc.) and HCAP cognitive categories	1	Statistical model Multivariate regression	MCI	Categorical status not validated Average of predicted probabilities matched HCAP prevalences very well. Strong association with education.
1.5 SDs below adjusted mean score [114,115,116] 1.5 SD below the mean adjusted (age/education) standardised score (registration, recall, fluency)	3	Standardised score SDs below mean	MCI	Not validated No normative sample used.
1.5 SDs below adjusted mean score [113] 1.5 SD below the mean adjusted (age/education) standardised score (registration, recall, fluency, serial7s)	1	Standardised score SDs below mean	MCI	Not validated No normative sample used.
CI criteria Description	Studies (n = 23) Criteria (n = 17)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Modified DemTect [54,55] Modified DemTect (registration, recall, verbal fluency, orientation, numeracy). Cut-off ≤ 14/20	2	CSI cut-off Selected cut-off	Poor cognitive functioning	Not validated Not adjusted for age/education.
1.5 SDs below adjusted mean [63] 1.5 SDs below mean for 34-point CSI (registration, recall, fluency, orientation) by country of residence	1	CSI cut-off SDs below mean	Cognitive impairment	Not validated No normative sample used. Not adjusted for age/education.
1 SD below adjusted mean [52] 1 SD below mean for a 29-point CSI (registration, recall, orientation, serial7s) by age	1	CSI cut-off SDs below mean	Ageing-associated cognitive decline	Not validated No normative sample used.
Cut-offs (≥1 domain) [99] Orientation < 3/4 or numeracy < 2/5	1	Cut-offs (domains) ≥1/2 cut-offs	Limited cognitive function	Not validated Not adjusted for age/education.
Cut-offs (2 domains) [101] (Also had number of impairments: 0, 1, 2) Both memory (registration < 5 and/or recall < 4) and fluency (<15)	1	Cut-offs (domains) Multiple cut-offs	Cognitive impairment	Not validated Not adjusted for age/education. Two domains typical of Alzheimer’s.
1.5 SDs below adjusted mean [117,118,119,120] Both memory (registration and/or recall) and fluency are 1.5 SDs below the mean by age	4	Cut-offs (domains) SDs below mean	Cognitive disorder; cognitive impairment	Not validated No normative sample used.
Cut-off/1.5 SDs below adjusted mean [121] Memory (20-word recall) 1.5 SDs below the mean by age and country (1 SD 75 years) or verbal fluency < 15	1	Cut-offs (domains) SDs below mean/cut-off	Cognitive impairment	Not validated No normative sample used for memory. Fluency not adjusted for age/education.
25th percentile (1 subtests) [16] [ATHLOS Project] One of the following tests in the lowest 25%/quartile (registration; recall; fluency)	1	Cut-offs (subtests) 25th percentile	Low cognitive functioning	Not validated Not adjusted for age/education. High cut-off for CI (false positives).
1.5 SDs below adjusted mean (≥2 subtests) [122] Two or more subtests (registration, recall, orientation) 1.5 SDs below mean by education.	1	Cut-offs (subtests) SDs below mean	Cognitive impairment	Not validated No normative sample used.
1.5 SDs below adjusted mean (≥2 subtests) [123] Two or more subtests (registration, recall, fluency orientation) 1.5 SDs below mean by education.	1	Cut-offs (subtests) SDs below mean	Cognitive impairment	Not validated No normative sample used.
Multiple selected cut-offs (≥3 subtests) [102] Three or more low scores (registration < 5/10, recall < 4/10, fluency < 15/100, orientation < 2/4, serial7s < 2/5).	1	Cut-offs (subtests) Multiple cut-offs	Cognitive impairment	Not validated Not adjusted for age/education.
1.5 SDs below adjusted mean [124] 1.5 SDs below mean in at least 1 subtest (registration, recall, fluency, orientation) by age	1	Cut-offs (subtests) SDs below mean	Cognitive condition	Not validated No normative sample used.
1.5 SDs below adjusted mean [125] 1.5 SDs below the mean standardised score (registration, recall, fluency) by education	1	Standardised score SDs below mean	Cognitive impairment	Not validated No normative sample used.
≤10% percentile [126] ≤10% percentile on standardised z-score (registration, recall, fluency, orientation, numeracy)	1	Standardised score percentile	Cognitive impairment	Not validated Not adjusted for age/education.
≤10% percentile [127] ≤10% percentile on standardised z-score (registration, recall, fluency, orientation, serial 7s)	1	Standardised score percentile	Impaired cognition	Not validated Not adjusted for age/education.
≤10% percentile by sex [128] ≤10% percentile by sex on standardised t-score	1	Standardised score percentile	Poor cognitive function	Not validated Not adjusted for age/education.
Machine learning with principal components [129,130,131] Unsupervised machine learning classification (hierarchical clustering on principal components). Both used registration and recall and one included orientation	3	Statistical model Machine learning	High likelihood of dementia	Used R packages FactoMineR, NbClust and missMDA.
Amnestic impairment criteria Description	Studies (n = 2) Criteria (n = 2)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Memory cut-offs [41] 1.5 SDs below mean 20-point recall score by age	1	CSI cut-off SDs below mean	Mild memory impairment	Not validated But multiple papers have applied the same cut-offs of <5 and <4, respectively.
Memory cut-offs [100] Registration < 5 or recall < 4	1	Cut-offs (subtests) ≥1/2 cut-offs	Memory impairment	Not validated But multiple papers have applied the same cut-offs of <5 and <4, respectively.

Bold text highlights key descriptive information and the shading marks the start of the results for each of the four cognitive classifications (probable dementia, CIND, MCI, and CI) to aid readability. Note while the total number of studies in the table sum to 40 (7 + 2 + 6 + 23 + 2) the total number of unique studies was 36 since four of the probable dementia studies also appear under CIND and the first two rows of MCI.

Table 5. Definitions of cognitive categories and stages identified in the SHARE studies classifying participants with probable dementia, cognitive impairment no dementia (CIND), mild cognitive impairment (MCI), and cognitive impairment (CI) defined as MCI or dementia.

Probable Dementia Criteria Description	Studies (n = 7) Criteria (n = 21)	Cognitive Measure Classification Approach	Terminology Used	Validation Status Notes
Langa–Weir criteria [33] Langa–Weir (registration, recall, serial 7s, counting backwards) ≤ 6/27	1	CSI cut-off Selected cut-off	Probable dementia	Validated in United States Best cut-off based on a gold standard diagnosis in the ADAMS [33].
Langa–Weir criteria (modified) [57] CSI (registration, recall, serial 7s, verbal fluency) ≤ 6/26	1	CSI cut-off Selected cut-off	CI consistent with dementia	Not validated Cut-off from Langa–Weir. (swapped 2-point counting with 1-point fluency).
Modified DSM-5 [38] Non-amnestic battery ≥ 2 SDs below mean by age/education (or self-reported dementia/memory disorder) with self-reported IADL difficulty (≥1 of telephone calls, taking medications, managing finances)	1	CSI cut-off SDs below mean + ADLs	Dementia	Not validated Cognitive battery did not include memory issues. IADL difficulty for cognitively orientated tasks only, not necessarily loss of independence. Prevalence was low.
LW algorithms (X2) [37] 20-point word recall 2.5% percentile (i.e., ≥2 SD below mean). Applied with or without IADL difficulties taking a cut-off of 1.5 interquartile ranges above Q3 (by country)	1	CSI cut-off Percentile + ADLs	Probable dementia	Percentile LW algorithms with ADLs recommended Cognitive battery only included memory issues. IADL difficulty had different cut-offs by country, not loss of independence. Performed well in this study.
1.5 SD below adjusted mean [51] 1.5 SD below 25-point CSI (registration, recall, serial7s) by age, education, sex, proxy status of interviews	1	CSI cut-off SDs below mean	Probable dementia	Not validated No normative sample used for memory. 1.5 SDs below a mean for MCI/CIND.
LW algorithms (X2) [37] Used 20-point word recall cut-offs from equipercentile with country prevalence values in OECD. Applied with or without IADL difficulties taking a cut-off of 1.5 interquartile ranges above Q3 (by country)	1	CSI cut-off equipercentile	Probable dementia	Percentile LW algorithms with ADLs recommended Cognitive battery only included memory issues. IADL difficulty had different cut-offs by country, not loss of independence.
Cut-off/1.5 SDs below adjusted mean [112] Memory (20-word recall) 1.5 SDs below the mean by age or verbal fluency < 15	1	Cut-offs (domains) SDs below mean / cut-off	Dementia	Not validated No normative sample used for memory. Not adjusted for age/education for fluency. 1.5 SDs below a mean for MCI/CIND.
1.5 SDs below mean [110] Standardised score (registration, recall, fluency, orientation) 1.5 SDs below mean and BADL difficulty (bathing, eating, dressing, transferring, or walking) or self-reported dementia	1	Standardised score SDs below mean + ADLs	Dementia	Not validated Not adjusted for age/education.
Random forest (X3) [37] Random Forest fitted using three training dataset approaches (registration, recall, ADLs)	1	Statistical model Machine learning	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
XGBoost classifier (X3) [37] XGBoost classifier in three training dataset approaches (registration, recall, ADLs)	1	Statistical model Machine learning	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Logistic regression model (X3) [37] logistic regression fitted to self-reported cases (registration, recall, ADLs)	1	Statistical model Regression	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Weighted logistic regression [37] Weighted logistic regression fitted to self-reported cases (registration, recall, ADLs)	1	Statistical model Regression	Probable dementia	Percentile LW algorithms with ADLs recommended Complex method but the authors provide R codes for it.
Predicted probabilities [33] (SHARE-HCAP) A predictive model was fitted between wave 9 items (cognition, ADLs, etc.) and HCAP cognitive categories	1	Statistical model Multivariate regression	Dementia	Categorical status not validated Average of predicted probabilities matched HCAP prevalences very well. Strong association with education.
CIND criteria Description	Studies (n = 2) Criteria (n = 2)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Langa–Weir criteria [33] Langa–Weir (registration, recall, serial 7s, counting backwards) 7–11 out of 27	1	CSI cut-off Selected cut-off	CIND	Validated in United States Best cut-off based on a gold standard diagnosis in the ADAMS [33].
Langa–Weir criteria (modified) [57] CSI (registration, recall, serial 7s, verbal fluency) 7–11 out of 26	1	CSI cut-off Selected cut-off	Probable dementia	Not validated Cut-off from Langa–Weir. (swapped 2-point counting with 1-point fluency).
MCI criteria Description	Studies (n = 6) Criteria (n = 4)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Modified Petersen’s criteria [38] Non-amnestic battery and self-reported questions on ADLs, memory, dementia used to define subjective memory complaints, MCI and dementia	1	CSI cut-off SDs below mean	MCI	Not validated MCI definition incomplete—missing amnestic MCI single domain.
Predicted probabilities [33] (SHARE-HCAP) A predictive model was fitted between wave 9 items (cognition, ADLs, etc.) and HCAP cognitive categories	1	Statistical model Multivariate regression	MCI	Categorical status not validated Average of predicted probabilities matched HCAP prevalences very well. Strong association with education.
1.5 SDs below adjusted mean score [114,115,116] 1.5 SD below the mean adjusted (age/education) standardised score (registration, recall, fluency)	3	Standardised score SDs below mean	MCI	Not validated No normative sample used.
1.5 SDs below adjusted mean score [113] 1.5 SD below the mean adjusted (age/education) standardised score (registration, recall, fluency, serial7s)	1	Standardised score SDs below mean	MCI	Not validated No normative sample used.
CI criteria Description	Studies (n = 23) Criteria (n = 17)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Modified DemTect [54,55] Modified DemTect (registration, recall, verbal fluency, orientation, numeracy). Cut-off ≤ 14/20	2	CSI cut-off Selected cut-off	Poor cognitive functioning	Not validated Not adjusted for age/education.
1.5 SDs below adjusted mean [63] 1.5 SDs below mean for 34-point CSI (registration, recall, fluency, orientation) by country of residence	1	CSI cut-off SDs below mean	Cognitive impairment	Not validated No normative sample used. Not adjusted for age/education.
1 SD below adjusted mean [52] 1 SD below mean for a 29-point CSI (registration, recall, orientation, serial7s) by age	1	CSI cut-off SDs below mean	Ageing-associated cognitive decline	Not validated No normative sample used.
Cut-offs (≥1 domain) [99] Orientation < 3/4 or numeracy < 2/5	1	Cut-offs (domains) ≥1/2 cut-offs	Limited cognitive function	Not validated Not adjusted for age/education.
Cut-offs (2 domains) [101] (Also had number of impairments: 0, 1, 2) Both memory (registration < 5 and/or recall < 4) and fluency (<15)	1	Cut-offs (domains) Multiple cut-offs	Cognitive impairment	Not validated Not adjusted for age/education. Two domains typical of Alzheimer’s.
1.5 SDs below adjusted mean [117,118,119,120] Both memory (registration and/or recall) and fluency are 1.5 SDs below the mean by age	4	Cut-offs (domains) SDs below mean	Cognitive disorder; cognitive impairment	Not validated No normative sample used.
Cut-off/1.5 SDs below adjusted mean [121] Memory (20-word recall) 1.5 SDs below the mean by age and country (1 SD 75 years) or verbal fluency < 15	1	Cut-offs (domains) SDs below mean/cut-off	Cognitive impairment	Not validated No normative sample used for memory. Fluency not adjusted for age/education.
25th percentile (1 subtests) [16] [ATHLOS Project] One of the following tests in the lowest 25%/quartile (registration; recall; fluency)	1	Cut-offs (subtests) 25th percentile	Low cognitive functioning	Not validated Not adjusted for age/education. High cut-off for CI (false positives).
1.5 SDs below adjusted mean (≥2 subtests) [122] Two or more subtests (registration, recall, orientation) 1.5 SDs below mean by education.	1	Cut-offs (subtests) SDs below mean	Cognitive impairment	Not validated No normative sample used.
1.5 SDs below adjusted mean (≥2 subtests) [123] Two or more subtests (registration, recall, fluency orientation) 1.5 SDs below mean by education.	1	Cut-offs (subtests) SDs below mean	Cognitive impairment	Not validated No normative sample used.
Multiple selected cut-offs (≥3 subtests) [102] Three or more low scores (registration < 5/10, recall < 4/10, fluency < 15/100, orientation < 2/4, serial7s < 2/5).	1	Cut-offs (subtests) Multiple cut-offs	Cognitive impairment	Not validated Not adjusted for age/education.
1.5 SDs below adjusted mean [124] 1.5 SDs below mean in at least 1 subtest (registration, recall, fluency, orientation) by age	1	Cut-offs (subtests) SDs below mean	Cognitive condition	Not validated No normative sample used.
1.5 SDs below adjusted mean [125] 1.5 SDs below the mean standardised score (registration, recall, fluency) by education	1	Standardised score SDs below mean	Cognitive impairment	Not validated No normative sample used.
≤10% percentile [126] ≤10% percentile on standardised z-score (registration, recall, fluency, orientation, numeracy)	1	Standardised score percentile	Cognitive impairment	Not validated Not adjusted for age/education.
≤10% percentile [127] ≤10% percentile on standardised z-score (registration, recall, fluency, orientation, serial 7s)	1	Standardised score percentile	Impaired cognition	Not validated Not adjusted for age/education.
≤10% percentile by sex [128] ≤10% percentile by sex on standardised t-score	1	Standardised score percentile	Poor cognitive function	Not validated Not adjusted for age/education.
Machine learning with principal components [129,130,131] Unsupervised machine learning classification (hierarchical clustering on principal components). Both used registration and recall and one included orientation	3	Statistical model Machine learning	High likelihood of dementia	Used R packages FactoMineR, NbClust and missMDA.
Amnestic impairment criteria Description	Studies (n = 2) Criteria (n = 2)	Cognitive measure Classification approach	Terminology used	Validation status Notes
Memory cut-offs [41] 1.5 SDs below mean 20-point recall score by age	1	CSI cut-off SDs below mean	Mild memory impairment	Not validated But multiple papers have applied the same cut-offs of <5 and <4, respectively.
Memory cut-offs [100] Registration < 5 or recall < 4	1	Cut-offs (subtests) ≥1/2 cut-offs	Memory impairment	Not validated But multiple papers have applied the same cut-offs of <5 and <4, respectively.

Bold text highlights key descriptive information and the shading marks the start of the results for each of the four cognitive classifications (probable dementia, CIND, MCI, and CI) to aid readability. Note while the total number of studies in the table sum to 40 (7 + 2 + 6 + 23 + 2) the total number of unique studies was 36 since four of the probable dementia studies also appear under CIND and the first two rows of MCI.

4. Discussion

This scoping review and instrument mapping exercise is the first study to identify different ways of measuring cognitive impairment in the SHARE and found that a large number of heterogeneous approaches have been used. Through the identification and synthesis of 234 SHARE studies using objective cognitive measures, we provide a novel overview of the wide range of methods applied to assess cognition, including the use of individual subtests, composite scores, validated and non-validated CSIs, statistical modelling approaches, and classification using diagnostic cut-offs. In addition, by mapping existing brief CSIs from the literature to the cognitive subtests available within the SHARE, we offer important insights into which instruments may be feasibly reconstructed or adapted for use in this longitudinal European cohort.

A large number of studies reported the results of subtests individually and/or limited the analysis to mean scores. The most common composite approach was to use single subtests to assess cognition (n = 94) followed by the development of CSIs for use directly within the SHARE, of which 22 unique instruments were identified, albeit only five were used more than once and just two were found to be validated: the recently published SHARE-Cog and the Langa–Weir Criteria. The Langa–Weir criteria was available but limited to a subsample of participants aged ≥ 60 years, while a new CSI, the SHARE-Cog, a bespoke three-item CSI incorporating the three most commonly found subtests in the SHARE across waves, was also identified [38] and was also available for ages 50–59 years. This is similar to how, in the HRS, the TICS [19] was limited to those aged ≥ 60 years and the Langa–Weir tool was generated to include those aged 50–59 years [42]. Comparisons between the SHARE-Cog and Langa–Weir tool are needed.

This study also assessed the feasibility of replicating established, published CSIs within this longitudinal ageing cohort by mapping available subtests and scoring methods available in the dataset compared to those available in the literature, examining the possibility of reconstructing these in the SHARE with high fidelity. It found that the subtests verbal fluency (for animals), working memory (i.e., 10-word registration) and episodic memory (i.e., 10-word delayed recall) were the three most commonly used items and were each used in over 75% of studies. This reflects the limited availability of individual subtests covering other domains across earlier waves, restricting the ability to develop replicable, widely used and validated short CSIs in the SHARE. This mapping exercise identified that other potential CSIs could be developed, though only a few had 100% fidelity with the original instruments. These included the 10-Point Cognitive Screener [70], the Six-Item Screener [72], and the Mini-Cog [71]; however, only the 10-point CS is available across historical waves but not fully. The Mini-Cog was limited to waves 8 and 9 and was only partially available. To date, none of these have been validated formally in the SHARE, and more research is needed to assess this.

Other more widely validated instruments such as the MoCA (67% fidelity of the original MoCA’s scoring items could be replicated using SHARE data) [18], Qmci screen (67% fidelity with the original) [85] and standardised MMSE (60% fidelity) [17,132] could be replicated in the SHARE, albeit with less than 100% fidelity.

As more subtests have been added across waves, including those examining visuospatial and executive function, the potential to develop more complex short CSIs targeting multiple cognitive domains and reflective of instruments in clinical practice has been realised. Reflecting that earlier waves measured only limited domains, primarily focused on the assessment of memory, the most common approach was to use the 10-word registration and recall subtests creating a 20-point score. This 10-word delayed recall question is also available in most other HRS-related studies [10,15], and the 20-point score has been used in other large longitudinal studies of ageing such as in the China Health and Retirement Longitudinal Study in China [133] and the English Longitudinal Study of Ageing in England [133]. Given its widespread use, 20-point recall score should be considered as a stand-alone subtest for future SHARE studies examining memory as it is consistent and comparable within the SHARE and across other longitudinal studies of ageing. In addition, it has been illustrated that the use of 10-word recall (as opposed to a smaller list of words to recall) improves the diagnostic performance when clinically differentiating between probable MCI and dementia [38]. It should be noted, however, that the overreliance on memory-based measures in the SHARE limits the ability to better detect non-amnestic profiles of impairment and, hence, the ability to detect non-AD related MCI or prodromal dementia syndromes.

For global cognition (including at least two cognitive domains), it was common to use verbal fluency (animal naming) in combination with word registration and recall. Verbal fluency has the greatest availability amongst possible non-amnestic tests in the SHARE (albeit limited to a small subsample in wave 7). Verbal fluency is also widely assessed across other studies [15], with the word recall tests and verbal fluency being selected by the Ageing Trajectories of Health-Longitudinal Opportunities and Synergies (ATHLOS) project which created a harmonised dataset to compare cognitive function across nine population-based studies [16]. Several options are available for combining these three scores, with one straightforward and validated approach being a weighted sum as in the SHARE-Cog instrument [38]. Other studies have used standardised (z-) scores to combine them or principal component analysis; however, these approaches may be more difficult to interpret. A z-score for a combined sum (such as the SHARE-Cog) could also be considered, and based on the z-score formula [92], providing the reference mean and reference standard deviation (SD) would allow for a conversion back to raw scores.

For identifying cognitive impairment and cognitive categories (stages), multiple approaches were proposed, and none of these have been well validated or used consistently across studies. A cut-off of 1.5 SDs below the sample mean is the most common approach to date; however, this remains an unvalidated approximation. In a normal distribution, 1.5 SDs below the sample mean is approximately equal to the 7th percentile. So, without isolating a normative subsample as detailed in the HCAP studies [13,134], this approach will always select approximately 7% of the lowest scores (assuming the CSI follows a normal distribution) regardless of the actual prevalence of CI. Thus the prevalence becomes statistically fixed. One study used a measure of ageing-associated cognitive decline defined as 1 SD below the mean score in each group [52], which is approximately the lowest 16% of participants in each age group (i.e., removes prevalence differences by age). Few studies mirrored “real world” clinical practice, with only one study using the well-established Petersen’s criteria for MCI, upon which the 2013 DSM-5 criteria for minor neurocognitive disorder are based [43]. Only a small number of studies adjusted for recognised confounders such as age or education, which is a key consideration in clinical practice. However, the utility of this in an epidemiological context is subject to some debate [135]. If adjusting for these variables, researchers should use an appropriate normative subsample as detailed in the HCAP studies to define age-education specific cut-offs [13,134], rather than the full sample. As stated above using the full sample (instead of a valid “normative” sample), to generate age–education-specific cut-offs would statistically remove any age/education prevalence differences (provided the CSI follows a normal distribution in each group). Further research is needed to compare and validate cognitive criteria in the SHARE. Further validation of cut-off scores is needed for both the 20-point word recall test and/or the SHARE-Cog.

Strengths of this study includes the breadth of the review. We searched both the SHARE study register and PubMed and employed citation tracking to identify additional eligible studies. By applying comprehensive eligibility criteria and following the PRISMA-ScR reporting framework [24], and a PRISMA2020 diagram to clearly document the sources of studies and increase replicability [25,26], this ensured methodological transparency. The inclusion of studies published in languages other than English further increases the comprehensiveness of the review, particularly for a multinational dataset such as the SHARE. The mapping of brief CSIs against SHARE subtests, conducted as a distinct second component of this study, enhances the utility of the findings by providing researchers with practical tools for standardising cognitive measures in future studies. In addition, the Supplementary Material (Table S2) documents all located studies for additional transparency. While the SHARE register provides a consistent and reliable source of SHARE studies our findings illustrate it is currently incomplete with almost 40% of included studies coming from PubMed and citation tracking. Another important contribution of this study is the identification of 24 CSIs from the broader literature that could potentially be reconstructed within the SHARE with at least 50% fidelity, including three instruments with complete fidelity, albeit incomplete availability across past waves. These findings are of practical relevance, particularly in light of the growing demand for brief and reliable cognitive tools that can be used consistently across ageing studies. The mapping exercise also offers a foundation for future validation work using SHARE data, including internal validation of CSIs for different age groups or across countries with varying cultural and educational profiles.

Limitations of this study include that this study focuses on cognition in the SHARE study only, rather than the wider context of HRS-related studies illustrated on the Gateway to Global Aging Data [10]. In addition, gender differences or other diseases were not assessed. Measures available in the SHARE should be harmonisable with most other HRS studies [15], although there can be subtle differences in measurement which can limit comparisons [39]. The SHARE itself is also a multinational cohort. Only a limited number of CSIs were identified from the systematic review papers that were published, and a systematic search of the literature to identify these was not conducted. While the two reviews were comprehensive and reflect the majority of commonly used short CSIs, some that could have been replicated in the SHARE may have been missed. Another limitation is that the data extraction and charting process was conducted primarily by a single reviewer, which may have introduced subjectivity or error in classification despite iterative discussions and consensus with a second reviewer. While the use of automated tools and structured forms helped mitigate this, a dual screening and data extraction process would have further enhanced reliability. In addition, the mapping of CSIs to SHARE subtests, although systematic, involved some degree of judgement, particularly where substitutions were made for similar but not identical subtests. While this pragmatic approach mirrors real-world research applications, it introduces variability that may influence the psychometric equivalence of the adapted instruments. Future research will be needed to empirically evaluate the diagnostic performance and validity of these mapped instruments within the SHARE. Similarly, as high structural fidelity does not automatically translate into psychometric or clinical validity in cohort studies such as the SHARE, i.e., that structural and conceptual equivalence do not always translate diagnostic equivalence, additional research in clinical studies is now needed to confirm that the reconstructed instruments theorised here are accurate, reliable and valid in clinical practice.

Finally, the review was not designed to assess the quality or risk of bias of the included studies, as is typical for scoping reviews. While this allowed for a broad overview of existing methods, it limits the conclusions that can be drawn about the robustness or validity of individual findings. However, this was not the primary aim of the study, and the value of this work lies in its descriptive synthesis and generation of a practical framework for measuring cognition in the SHARE.

We note that some of the newly available cognitive subtests in the SHARE, such as clock drawing and cube copying, were not represented in many of the included studies. This reflects the recency of their introduction (wave 8 onward) but also suggests that their potential for assessing visuospatial or executive function remains underutilised. As such, future studies should explore whether the inclusion of these subtests improves the diagnostic performance of CSIs or composite scores in the SHARE. Moreover, while we identified a wide range of statistical approaches to measuring global cognition (e.g., z-scores, principal component analysis, latent growth models), few of these were validated against external gold standards, limiting their interpretability in clinical or epidemiological contexts.

To assess memory, we recommend using and including a summary score that combines both 10-word registration and recall, such as a 20-point scale. Common cut-offs include fewer than 5 words for registration and fewer than 4 words for recall. However, further research is needed to verify these thresholds and to establish cut-offs for the combined score. It may also be worthwhile to explore whether a weighted sum could improve diagnostic accuracy. For assessing global cognitive function, the combination of 10-word registration, recall, and verbal fluency is widely available. One option is the 45-point SHARE-Cog instrument, but further validation of this CSI and its cut-offs is necessary.

For identifying cognitive impairment, more research is needed to develop a method aligned with Petersen’s criteria for MCI and the more-recent DSM-5 criteria for minor neurocognitive disorders, which uses the principles outlined by Petersen. In the meantime, researchers could use either the Langa–Weir scale cut-offs or scores 1.5 SDs below the age- and education-adjusted mean. In recent epidemiological studies, the Langa–Weir algorithm for probable dementia has been proposed [37]. This method incorporates 20-point word recall, and limitations in ADLs and adjusts recall cut-offs using external dementia statistics. It has shown better performance than some sophisticated machine learning models and is available for all major SHARE waves, including wave 7. However, it lacks measures of non-amnestic cognitive impairment and depends on the accuracy of external prevalence data. For MCI and dementia classification, a regression model based on HCAP data has also been proposed [33], which estimates the probability of MCI or dementia rather than assigning a categorical diagnosis. This holds promise for future waves of the SHARE although the model may not be reliable for those aged < 65 years who were not included in HCAP.

The present study focuses on identifying widely available CSIs, and these may be used for identifying cognitive impairment in an efficient and consistent manner. Case-finding and opportunistic cognitive screening of high-risk individuals followed by more detailed neuropsychological assessment will remain an important clinical approach. However, more recent approaches such as combining neuropsychological assessments with other plasma and cerebrospinal fluid neurodegenerative biomarkers and other objective biomarkers, like measures of intracortical connectivity with transcranial magnetic stimulation, show much promise for even earlier detection [136], and a clear validation framework has been proposed [137]. In light of widely applicable non-invasive biomarkers (such as those for frontotemporal dementia [138]) the role of these CSIs in clinical care may shift somewhat. Hence, they will remain important epidemiological tools for monitoring changes in prevalence and incidence and may still be used to quantify cognitive decline over time as well as to help differentiate MCI subtypes [136].

5. Conclusions

This study provides the first review of cognitive measurement approaches, instruments and subtests within the SHARE study, revealing a high degree of methodological heterogeneity and limited use of validated CSIs. Through a structured instrument mapping process, we identified several brief CSIs that can be reconstructed using SHARE subtests, with some achieving full fidelity. Multiple unique CSIs have been utilised in SHARE to date with little validation/consistency, and a large number of potential cognitive screening approaches were identified that could be used. The inclusion of additional subtests in later waves and the development of the SHARE HCAP will also enable more comprehensive assessments. To support longitudinal analyses of cognitive change in the SHARE, and given the availability across waves, we recommended combining at a minimum the word registration and recall subtest with verbal fluency into a single instrument to provide a broad measure of cognition including several cognitive domains. The recent internal validation of the SHARE-Cog affirms this and represents a significant step forward, but further research is needed to harmonise and standardise cognitive assessment in the SHARE. For cross-national comparisons, instruments with consistent item availability and harmonised scoring are essential, with the SHARE-Cog or similarly constructed composites offering a pragmatic balance between validity and comparability, while internally standardised or percentile-based cut-offs should be used cautiously. For prevalence estimation and classification of cognitive impairment or dementia, approaches with external validation are critical; currently, the Langa–Weir algorithm and HCAP-derived probabilistic models provide the strongest epidemiological grounding, particularly when cognitive performance is combined with functional impairment. Aligning measurement choice with analytic purpose is therefore essential to improve validity, comparability, and interpretability of findings derived from SHARE data.

In conclusion, these findings offer an important resource for researchers and clinicians seeking to use SHARE data to investigate cognitive ageing and impairment and underscore the importance of consistency, transparency, and psychometric validation in the use of cognitive measures across international ageing studies.