Accuracy of the Informant Questionnaire on Cognitive Decline in the Elderly for Detecting Preexisting Dementia in Transient Ischemic Attack and Stroke

Supplemental Digital Content is available in the text. Background and Purpose: Prestroke dementia prevalence is high and impacts outcome. Although the IQCODE (Informant Questionnaire on Cognitive Decline in the Elderly) is being used to assess prestroke cognition, data on its validity for prestroke dementia are lacking. We studied the accuracy of the short-form (16-item) IQCODE for pre-event dementia in a population-based study of all transient ischemic attack (TIA)/stroke. Methods: All patients with TIA/stroke in a defined population of ≈92 720 (Oxford Vascular Study, 2002–2017) with IQCODE were included. IQCODE questionnaires were given to participants at baseline interview with instructions to pass to an informant for completion and return by post. Diagnosis of pre-event dementia was defined as prior diagnosis of dementia, or dementia by the Diagnostic and Statistical Manual of Mental Disorders-IV criteria on study interview and hand-searching of the entire medical record blinded to IQCODE. Reliability of the IQCODE for dementia was determined by the area under the receiver operating characteristic curve, sensitivity and specificity, stratified by age, event severity, and first-ever stroke. Results: Among 2059 interviewed survivors, IQCODE were returned in 1068 (mean age/SD=72.9/12.3, 47% TIA, 52.3% male, 68 [6.4%] pre-event dementia). Area under the receiver operating characteristic curve for IQCODE for pre-event dementia was 0.94 (95% CI, 0.90–0.97, P<0.001) with similar results by age: 0.92, 0.88 to 0.96, <65 years; 0.94, 0.83 to 1.00, 65 to 74 years; 0.95, 0.92 to 0.99, 75 to 84 years; 0.89, 0.82 to 0.96, ≥85 years. The optimal cutoff score overall was >3.48 (sensitivity=89.7%; specificity=84.2%) but was nonsignificantly higher for major stroke (National Institutes of Health Stroke Scale score ≥3) than minor stroke/TIA (>3.85 versus >3.47). Performance was similar in patients with first-ever stroke (area under the receiver operating characteristic curve, 0.92 [0.88–0.97]; sensitivity=85.7%; specificity=84.8% for cutoff >3.48). All 16-IQCODE questions discriminated between dementia and no dementia (all P<0.001) with the greatest differences seen for finances, using gadgets, arithmetic, and learning new things. Conclusions: IQCODE has excellent accuracy for detecting preexisting dementia in TIA and stroke with the pattern of deficits suggesting prominent executive dysfunction.

C ognitive impairment is associated with increased risk of stroke and of more severe stroke. 1,2 The prevalence of prestroke dementia ranges from 5% in transient ischemic attack (TIA) to 21% in severe major stroke (National Institutes of Health Stroke Scale score >10). 1 Prestroke cognitive decline also increases the risk of death from stroke and of poststroke dementia. [3][4][5][6] However, identifying pre-TIA/stroke cognitive status at the time of the index cerebrovascular event may be difficult. Although some patients will have a documented dementia diagnosis, rates of undiagnosed dementia are high: around one-half of hospitalized people aged ≥75 years have moderate/severe cognitive impairment but only around 20% have a diagnosis of dementia. 7,8 Preexisting dementia can be excluded in patients with good cognitive function after TIA/stroke but for cognitively impaired patients, a retrospective assessment of cognition will be required. In such cases, indirect assessment via relatives or caregivers is necessary facilitated by structured tools including the IQCODE (Informant Questionnaire on Cognitive Decline in the Elderly), Eight-item Informant Interview to Differentiate Aging and Dementia (AD-8), and the general practitioner assessment of cognition, the latter being used in primary care settings. [9][10][11][12][13][14] The IQCODE measures global cognitive decline across a range of areas and is largely unaffected by occupation or education. 9,10,12 It was developed to assess community-dwelling older adults, in whom Alzheimer disease is the most common dementia subtype and amnestic deficits are prominent. Each IQCODE item is scored on an ordinal 5-point scale score (1)(2)(3)(4)(5), and the average score across all items is calculated. Higher scores indicate cognitive decline. There is no consensus on the optimal cutoff for dementia which may vary according to the population characteristics and as with all cognitive screening tests, there is a trade-off between sensitivity and specificity. 10,12 The IQCODE is available as 16-and 26-item versions that have similar validity and high internal consistency and test-retest reliability in nonstroke settings. 10,12 However, previous validation studies had possible selection bias and limited reporting of IQCODE application, missing data, and incomplete questionnaires. [10][11][12] There are also few data on differential item patterns across dementia subtypes.
The IQCODE has not been validated for prestroke dementia in cerebrovascular populations 15 although it has been used to quantify or exclude pre-event dementia in stroke studies using variable cutoffs (3.30, 3.35, 4.00). 4,5,[16][17][18][19][20] We, therefore, assessed the accuracy of the IQCODE for pre-TIA/stroke dementia, and the effect of event severity and informant characteristics, in a population-based observational study of all TIA and stroke.

Data Availability
Any requests for data will be considered by P.M. Rothwell (peter. rothwell@ndcn.ox.ac.uk).

Study Background
This study was conducted within the ongoing OXVASC (Oxford Vascular Study, 2002-) a longitudinal, prospective, population-based incidence study of all acute vascular events in a defined population of ≈92 720 in 9 primary care practices with around 100 general practitioners in Oxfordshire, United Kingdom. 21,22 OXVASC is approved by the local Oxfordshire clinical research ethics committee (CO.043). Written consent is obtained at inclusion for study interview and follow-up, including review of all primary care and hospital records and death certificates. Where a patient is unable to give consent, assent from a relative is obtained.
Patients with TIA or minor stroke are referred by their primary care physician or the emergency department to the daily OXVASC emergency clinic staffed by a dedicated OXVASC clinical team. Patients with major stroke are admitted to the regional acute stroke unit at the John Radcliffe Hospital, Oxford, covering the study population area and are seen as soon as possible after admission by a member of the study team. Ascertainment of TIA and stroke approaches 100% of events reaching medical attention. 21,22 Diagnosis and management are reviewed by the same senior neurologist (P.M. Rothwell) in all patients. Major stroke is defined as National Institutes of Health Stroke Scale score of 3 or greater, as this corresponds broadly to the inpatient versus outpatient population. 23

Participants
The current study included consecutive patients recruited during 2 periods when the IQCODE was included in the OXVASC study protocol (April 1, 2002, to February 31, 2004;March 31, 2010, to March 31, 2017. All patients with a stroke (ischemic and hemorrhagic) or definite/probable TIA, as adjudicated by P.M. Rothwell, were included.
Baseline interview was done using a structured proforma which included: medical history, risk factors, family history, medication, functional, and cognitive assessment. Reasons for lack of interview or cognitive test were documented. When a patient could not be interviewed (eg, because of aphasia, delirium, or dementia), a relative or caregiver was interviewed instead. Brain and vascular imaging were done at baseline.
The validated 16-item short IQCODE in English was given to the patient during the baseline interview. 10

SCIENCES
instructed to ask someone they had known for at least 10 years to complete the form, and post it back. Patients were free to choose any informant. Informants were asked to score cognitive function over the last 10 years up to just before the index event.
Each item was scored on a 5-item scale: 1: much improved, 2: a bit improved, 3: not much change, 4: a bit worse, 5: much worse. Informants or patients were called by a research nurse if the IQCODE was not returned within 2 weeks, or if they had very low scores or missing items. From 2010 onwards, the IQCODE included questions about the relationship of the informant to the patient. Pre-event dementia diagnosis was made as described previously 1 (and see also methods in the Data Supplement) using the following information blinded to the IQCODE assessment: (1) baseline clinical assessment by study clinician; and (2) any dementia diagnosis, and related consultations and investigations, where available, in the primary care record, with hand-searching of the entire record including individual consultations, clinic letters, and hospitalization documentation. 1 The diagnosis of pre-event dementia was made by a senior study physician with expertise in dementia (S.T. Pendlebury) using the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV criteria. 1 Dementia subtype diagnoses were recorded where available (eg, in the primary care or hospital records or clinic letters).

Statistical Analysis
Demographic and clinical characteristics were compared for patients with versus without IQCODE, using χ 2 or ANOVA as appropriate. We included IQCODEs with missing items and calculated mean item scores by dividing by the number of completed items.
Performance of individual IQCODE items was evaluated by calculating the difference in mean item scores of patients with versus without dementia and statistical significance was tested with ANOVA. Items were subsequently ranked in order of the differences in mean item scores between subjects with versus without dementia.
Validation of the IQCODE for pre-event dementia was done according to STARDdem criteria. 24 Reliability was determined using the area under the receiver operating characteristic curve (AUROC). Optimal cutoffs were calculated using Youden's index. 25 Sensitivity, specificity, positive, and negative predictive values were determined. AUROC, accuracy, and mean item scores were stratified by age, event severity, and dementia subtype (Alzheimer versus vascular), and subgroups were compared using ANOVA. We evaluated the differences between AUROCs with Z tests. 26 We also compared mean item scores and AUROC between patients who had the IQCODE but who were untestable with a directto-patient cognitive test (abbreviated mental test score We performed sensitivity analyses omitting IQCODEs with missing items and where mean item scores were <3.00 (indicating cognitive improvement over 10 years). To exclude any possibility of incorporation bias, that is, that IQCODE had been used in the pre-event dementia diagnosis, we also performed analyses including only those patients with a dementia subtype diagnosis since all such patients had received their dementia diagnosis from a specialist other than S.T. Pendlebury. Statistical significance was set at 0.05 and 95% CIs were derived as ±1.96× the SE of the estimate. All analyses were done in SPSS for Windows 25.0.

RESULTS
Of 2400 patients (mean age/SD=73.6/13.6 years, 1180 female) with a definite or probable TIA or stroke over the 9 year study period, 2248 survived to ascertainment, of whom 2059 (92%) were interviewed and given an IQCODE to pass on to a relative or carer to complete and return by post. Questionnaires were returned in 1068 (52%; Figure and Figure). However, AUROCs for preexisting dementia were similar in testable and untestable patients: 0.93 (0.90-0.97) and 0.93 (0.87-1.00), respectively.
Mean item scores in incomplete IQCODEs were comparable to those of fully completed questionnaires (3.28 versus 3.23, P=0.52). Sensitivity analyses showed that test accuracy was unaffected by removal of IQCODEs with missing items: sensitivity=88.9% and specificity=84.2% for cutoff of >3.48. The 117 patients with mean IQCODE score below 3.00 (indicating cognitive improvement over the last 10 years) were younger than those with mean scores ≥3.00: mean age/SD=68.6/12.6 years versus 73.4/12.1 years (P<0.001). Excluding these 117 patients did not significantly change accuracy: sensitivity=91.0%, speci-ficity=82.1% for cutoff of >3.48. Among 68 pre-event dementia cases, 40 had a prior diagnosed dementia subtype. To rule out possible effects of incorporation bias, we performed a further sensitivity analysis excluding dementia patients without a documented dementia subtype. Accuracy of the IQCODE did not significantly change: cutoff >3.48 had 92.5% sensitivity and 84.0% specificity.
All 16-IQCODE items differed significantly for patients with versus without pre-event dementia both overall and after stratification for event severity (all P<0.001; Table 3 and Table II in the Data Supplement). Items 14-finances, 9-gadgets, 15-arithmetic, and 10-learning new things had the biggest mean differences between those with versus without pre-event dementia. The best discriminating item 14-handling financial matters alone had AUROC of 0.89 ( Figure I in the Data Supplement). When stratified by event severity, major stroke showed greatest differences between dementia versus no dementia in items that might be considered as requiring greater executive function (items 14,9,15,10), whereas TIA and minor stroke showed a more mixed pattern (items 9, 15, 14, 10/2/3; Table II in the Data Supplement).

SCIENCES
Among the 40 patients with known dementia subtypes, there were 17 (42.5%) with Alzheimer disease, 16 (40.0%) with vascular dementia, 3 (7.5%) with Lewy Body disease or Parkinson dementia, 2 (5.0%) with mixed dementia, and 2 (5.0%) with other. There were no significant differences in mean item scores across dementia subtypes (P=0.50). Although there were no significant differences in question performance between vascular and Alzheimer disease subgroups, Alzheimer patients were worse at remembering things about family and in doing arithmetic whereas vascular dementia patients were worse in handling money, dealing with finances, and making decisions (Table 4).

DISCUSSION
The IQCODE had excellent accuracy for identifying preevent dementia with an overall optimal cutoff of >3.48. Only 3% returned IQCODES were incomplete, and of these, two-thirds had only 1 item missing. Missing items did not affect test accuracy. Patients with IQCODE who were untestable with a cognitive test were older and had more dementia suggesting that the IQCODE provided cognitive information in otherwise untestable patients and, therefore, mitigates selection bias. The pattern of deterioration across individual items suggested executive-type deficits especially in those with major stroke, in keeping with dementia with a vascular component.
To our knowledge, this is the first study to validate the IQCODE against a reference standard of prior clinical diagnosis of pre-TIA/stroke dementia. The original IQCODE validation found an optimal cutoff >4.0 with sensitivity of 92.7% and specificity of 88.1% but compared moderate/severely affected Alzheimer patients with community volunteers and was at risk of bias. 9 The cutoff of >4.0 has subsequently been used to exclude preexisting dementia in stroke studies 4,5,16-20 but our findings suggest that this will have resulted in some cases of dementia being included. The higher optimal cutoff we observed for major stroke than for TIA or minor stroke may be the result of more advanced/more severe dementia in those presenting with major stroke. 1 Just over a tenth of IQCODEs had a mean item score <3.00, indicating cognitive improvement. Studies in nonstroke populations rarely found low scores since aging is usually

CLINICAL AND POPULATION SCIENCES
associated with stable or declining cognition. 27 Mistakenly scoring early recovery after stroke as improvement might explain these low scores. Patients with low scores were younger, and young patients are more likely to improve after stroke. 28 IQCODE items about finances, money, arithmetic problems, and learning new things discriminated best between TIA/stroke patients with and without preexisting dementia but in previous nonvascular validation studies, memory items were also important. 9,10 There have been few studies on IQCODE-item performance across different dementia subtypes. One study reported that executive items showed differences between frontotemporal dementia and Alzheimer disease. 29 However, none of the 16 items differentiated between Alzheimer disease and vascular dementia subtypes in  CLINICAL AND POPULATION SCIENCES our study, possibly because of small numbers although there were some qualitative differences: patients with Alzheimer disease showed the biggest mean score difference in items associated with memory whereas vascular dementia patients scored worst on items requiring executive function. Informants were most often spouses (60%-65%) or children (27%-30%) in keeping with the few studies that have reported informant details. [30][31][32] The IQCODE was completed and returned in just over half of interviewed patients which is comparable to previous studies: lack of suitable informants may affect 12% to 40% inpatients. [31][32][33] Since poor social interaction is associated with higher dementia incidence, this group might be at high risk resulting in a lower measured dementia rate in those with IQCODE versus the population as a whole. 34 Overall, the IQCODE would appear a valid tool for use in stroke research to exclude patients with prestroke dementia in studies of the prevalence and predictors of poststroke dementia. The IQCODE may also be useful in clinical practice to guide rehabilitation strategies: for example, patients with preexisting dementia may not benefit from cognitive rehabilitation and could be identified using an IQCODE score favoring specificity over sensitivity to ensure that patients without dementia were not excluded inappropriately. The IQCODE may also provide clinically useful prognostic information: patients with prestroke dementia may be more likely to have a substantial Alzheimer disease component and, therefore, ongoing poststroke decline than those with poststroke dementia and normal IQCODE who may show a more fixed deficit pattern with slower progression. Similarly, patients with abnormal IQCODE not reaching the threshold for dementia are at greater risk of future dementia after stroke as shown in previous studies 5,6,35 although it remains unclear whether this is independent of baseline cognitive test score. In terms of the implications for clinical practice, the IQCODE findings may help in counseling patients and carers regarding the cognitive prognosis, likelihood of recovery and response to rehabilitation. In addition, studies in Alzheimer disease suggest that cognitive decline may be slowed by treatment of vascular risk factors and the acute TIA/stroke event provides an opportunity to properly address these through robust secondary prevention measures to reduce both the risk of future cerebrovascular events and dementia. 36 Strengths of our study include the large sample, population-based design, and limited exclusion criteria. We also performed a number of sensitivity analyses including for the effects of possible incorporation bias. Limitations include the fact that, despite the robust study design, the IQCODE was completed and returned less often in inpatients with more severe stroke and previous stroke in whom rates of pre-event dementia are higher, resulting in a lower rate of pre-event dementia (6.4%) in those with IQCODE than in the cohort overall (9.8%). 1 We did not specifically examine feasibility, cost, and acceptability in our study, but available data from elsewhere suggest that these are favorable. 12 In addition, in keeping with previous studies, we analyzed the IQCODE as a nominal mean item score which assumes that the difference between each item is equal, when in fact it is an ordinal 5-point scale. [10][11][12] We did not screen informants for cognitive problems or depression, anxiety, stress, or carer burden 11,31 or stipulate any requirement regarding the amount of time informants should spend in the company of the patient. 33 Finally, informant characteristics were only available for later subjects in our study.
In conclusion, the IQCODE has high accuracy for identifying pre-TIA/stroke dementia, with optimal cutoff >3.48. However, the cutoff used in practice will vary according to whether greater sensitivity or specificity is preferred, and whether major stroke versus minor stroke or TIA is being assessed. Further studies are required to examine the patterns of item performance in different dementia subtypes, the impact of informant characteristics, and the predictive value of the IQCODE for poststroke cognitive decline.