赞比亚成年人社区痴呆筛查访谈(CSI-D)的因素结构
Factor structure of the Community Screening Interview for Dementia (CSI-D) among Zambian adults
赞比亚一项针对55岁及以上社区居民的横断面研究中,研究人员使用确认性因素分析评估了由经过培训的社区卫生工作者施行的社区痴呆筛查访谈(CSI-D)参与者访谈部分的理论推导因素结构。该研究旨在验证CSI-D在赞比亚这一低资源环境中的测量结构,以支持非专业人员开展痴呆筛查。
Introduction
Alzheimer’s Disease and Alzheimer’s Disease Related Dementias (AD/ADRD) affect over 51 million people worldwide, with two-thirds of those living in low- and lower-middle income countries (LMICs) like Zambia1,2. By 2050, the number of Zambian adults aged 60 and over is expected to increase to 3 million due to ongoing population growth as well as reductions in early childhood mortality3. The Zambian government has recognized the need for improved care for aging adults and, in 2015, introduced the National Policy on Aging4. Yet, informal caregivers continue to meet most care needs in Zambia and the number of healthcare providers with expertise in conditions that affect aging adults, particularly dementia, remains limited5.
Like several LMICs, Zambia has few neurologists and neuropsychologists6. Currently there are 18 Zambian neurologists and 39 neuropsychologists for over 21 million people. This expertise is concentrated at tertiary care centres in the capital city, Lusaka, whereas 60% of Zambians age 55 years and above reside in rural regions7. Dementia prevalence in Zambia is largely unknown and may range from 4.6%, as in nearby Tanzania, to as high as 27%, as shown in one rural hospital-based study8,9. Dementia diagnosis in Zambia is limited by lack of access to neurocognitive expertise and is likely further compounded by limited community awareness of dementia. While there is a dearth of data regarding community knowledge and perceptions of dementia in Zambia, studies from nearby South Africa and Democratic Republic of Congo have found that cognitive impairment is highly stigmatized and often associated with misconceptions related to witchcraft10,11. Thus, increasing access to health care workers that can diagnose dementia, provide disease education, and address stigmatizing beliefs will improve quality of life for Zambians as they age.
Community Healthcare Workers (CHWs) offer a valuable opportunity to transform dementia management. CHWs are an extension of formal health facilities’ workforce and provide education and basic health services in response to emerging health needs12. They are well known in their communities and, as a result, are trusted sources of health information. In sub-Saharan Africa, CHWs have played an integral role in increasing access to HIV services13. CHWs have also conducted hypertension screening and outreach in South Africa and have aided with community-based dementia research in rural Tanzania and South Africa14,15,16. Equipping CHWs with dementia knowledge and appropriate screening tools for cognitive impairment may improve access to dementia care, expand community awareness, and address community misconceptions that often accompany these disorders in LMICs.
Unfortunately, there are a limited number of culturally adapted, education-fair instruments validated to support dementia screening and diagnosis in LMICs. Options are even fewer for non-specialized practitioners like CHWs with limited access to technology. While traditional screens for dementia, such as the locally adapted Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA), are used clinically in Zambia, these tests are heavily weighted on skills learned through school exposure and, as a result, have limited utility for populations with low formal education or literacy17,18,19,20. The Community Screening Instrument for Dementia (CSI-D) was developed to assess dementia in two populations with disparate cultural and linguistic practices and subsequently validated for use in five settings21,22,23. It is an integral part of the 10/66 Dementia Protocol, which conducted population-based dementia studies in ten LMICs across multiple languages and cultures, including Nigeria22,23,24. The CSI-D includes a participant interview, which is comprised of a series of items assessing neurocognitive domains, as well as an informant interview. The CSI-D has been used across sociocultural contexts to assess dementia and a brief version has been validated for use by non-specialists health care providers25. The CSI-D offers an ideal starting point for identifying measures that non-physician healthcare providers, such as CHWs, can use to screen for dementia. While the 10/66 Dementia Protocol and the CSI-D have been widely used, and items were initially selected from existing dementia assessment tools21, there is limited research on the measurement structure of the CSI-D participant interview. Most studies have utilized the CSI-D’s existing sum scores, which includes a cognitive scale (cogscore) score and an informant (relscore) score created from participant and informant interview responses, respectively, and a discriminant function score (dfscore), which is a weighted combination of the cogscore and relscore, to assess dementia prevalence16,26,27,28. A few have compared dementia designation using the CSI-D cognitive or discrimination function score cutoffs with other dementia measures29,30, yet, it is not known whether the CSI-D participant interview in its current format reasonably summarizes the cognitive domains affected by dementia across populations. If the CSI-D participant interview does not adequately assess the cognitive domains as intended in Zambia, then it would require modification prior to widespread use our setting. Therefore, in addition to assessing the feasibility of implementing the CSI-D as a community-based screening tool for dementia in an LMIC, further evaluating the measurement structure of the CSI-D participant interview is also warranted to improve appropriate use.
In this study, trained CHWs administered the CSI-D participant interview as part of a cross-sectional, population-based study of community-dwelling Zambians aged ≥ 55 years in Lusaka. Confirmatory factor analysis was used to evaluate the theoretically derived factor structure of the CSI-D participant interview in relation to established cognitive domains.
Results
Study participants
Seven-hundred thirty-four adults ≥ 55 years across four Lusaka communities participated in this study (Table 1). Study participation was discussed with an additional 37 participants and 4 informants who declined participation in the parent study (response rate: 734/775, 95%). Most participants were female (502, 68%), with a mean age of 66.9 (SD 8.1) years, though notably, participant age ranged from 55 to 101 years. Formal educational attainment ranged from zero to 20 years, with a mean of seven years of education. Further, 384 (52.3%) completed primary school, whereas 105 (14.3%) completed secondary. The majority (560, 76%) were born outside of Lusaka, and the average age at which they moved to Lusaka was 25 years. The mean monthly household income was 74.6 (SD 92.4) United States Dollars (USD). Across the study communities, there were significant differences in demographic characteristics. Specifically, Chilenje participants were more likely to be female (75% vs Chawama 73%, Chipata 61%, and Matero 63%, p < 0.01), older (68.1 yrs vs. Chawama 66.3 yrs, Chipata 67.4 yrs, and Matero 65.4 yrs p < 0.01), and were less likely to be born outside of Lusaka compared to the other study communities (69% vs. Chawama 75%, Chipata 87%, and Matero 76%, p < 0.01). Chilenje participants also had a higher level of formal education (8.4 yrs vs. Chawama 6.3 yrs, Chipata 5.3 yrs, and Matero 7.5 yrs, p < 0.01) and greater average monthly household income (USD 103.2 vs. Chawama USD 52.8, Chipata USD 58.1, and Matero USD 75.7, p < 0.01).
Participant responses to CSI-D interview items
All 734 participants responded to each CSI-D participant interview item (Table 1). Two items had notably low correct response rates at 39% (“What is the name of the current member of parliament?”, 288/734) and 40% (intersecting pentagons, 291/734). Five items with the highest correct response rates were the following: “What do you do with a hammer?” (725/734), “What is a bridge?” (721/734), “What do people do when they go to church?” (729/734); “What is the name of the first president of Zambia?” (724/734), and “What is the name of the city we are in?” (725/734). With 98–99% correct, these items had limited variability in responses and thus exhibited a ceiling effect. Ninety-five percent of participants got all three-words correct on immediate recall, though only 42% repeated all three words correctly after a delay. On the modified CERAD, the mean number of words correct was 12.9/30 on immediate recall across three trials with an average of 4/10 words correct after a delay. Seventy-seven percent were able to name all objects shown to them. The objects with the fewest correct responses were a knuckle (609/734) and an elbow (661/734). Notably, items within the CSI-D participant interview had significantly different response rates between study communities (p < 0.05) (Table 1). The only items were not significantly different between communities were “What do you do with a hammer?” (p = 0.50), “What is a bridge?” (p = 0.06), “What do people do when they go to church?” (p = 0.19), and “What is the name of the city we are in?” (p = 0.90), all of which most participants answered correctly.
CSI-D participant interview dimensionality and item performance
Four cognitive domains were fit using the CSI-D participant interview items: memory, orientation, language, and visuospatial.
Memory
We found the cognitive domain of memory had satisfactory fit, with a CFI of 0.99 RMSEA of 0.02, and SRMR of 0.03 (Model 1, Table 2). We included residual correlations between the following items: repeat of interviewer’s name & recall of interviewer’s name; immediate and delayed three-word recall; and the modified CERAD immediate and delayed recall. Standardized factor loadings ranged from 0.43 (repeat interviewer’s name) to 0.64 (modified CERAD delayed recall). The item thresholds ranged from -1.64 (immediate three-word recall) to 0.21 (delayed three-word recall), and the continuous items had intercepts of 3.96, 5.03, and 12.93 for modified CERAD delayed recall, brave man story, and modified CERAD immediate recall, respectively (Model 1, Table 3).
Orientation
In the cognitive domain of orientation, we found adequate model fit: the CFI was 0.98, RMSEA was 0.05, and SRMR was 0.08 (Model 2, Table 2). The standardized loadings for the orientation domain ranged from 0.57 (“What is the name of the member of parliament?”) to 0.84 (“Where is the market?”). Further, the item thresholds ranged from −1.71 (“Where is the market?”) to 0.27 (“What is the name of the member of parliament?”) (Model 2, Table 3). Due to the conceptual and wording overlap between the items “What month is it?” and “What year is it?”, we included a residual covariance to the model. Additionally, the modification index was high for the addition of this covariance, 10.91, suggesting improved model fit with the residual covariance.
Language
The language model had satisfactory fit, as CFI was 0.98, RMSEA was 0.03, and SRMR was 0.07 (Model 3, Table 2). Due to the similarities in the one and two-step verbal commands “Nod your head” and “Point to the window then the door”, we specified a residual covariance for those respective items. The lowest standardized loading was 0.48 for the adapted phrase repetition and the highest loading was 0.84 for the command “Nod your head”. The one and two step verbal commands had the lowest item thresholds, −1.99, while the threshold for getting all seven shown items correct (Shown items | 2) was the highest, −0.75 (Model 3, Table 3).
Visuospatial
The single factor CFA model for visuospatial cognitive domain only had two items, drawing intersecting circles and drawing intersecting pentagons. Thus, this model was fit by forcing the two item loadings to be equal to one, resulting in fit indices that are perfect (CFI = 1.00, RMSEA = 0.00, and SRMR = 0.00) (Model 4, Table 2). These metrics indicate perfect fit because of limited degrees of freedom and items available in the visuospatial domain. The thresholds, however, for drawing intersecting circles was −0.83, and for intersecting pentagons was 0.26 (Model 4, Table 3).
CSI-D participant interview items removed
Five items from CSI-D participant interview were removed from all CFA models: (1) “What do you do with a hammer?”; (2) “What is a bridge?”; (3) “What do people do when they go to church?”; (4) “What is the name of the first president of Zambia?”; and (5) “What is the name of the city we are in?”. These items had the lowest variability in response rates, with correct response rates ranging from 98% to 99% (Table 1). Further, upon inspection, these questions had low correlations with other items, and when implemented in first-order CFA models had loadings below 0.40. Thus, these questions were removed from CFA models and discussed descriptively.
Second-order CFA model
We then fit a second-order factor model to evaluate the relationship between the first-order domains memory, orientation, language, and visuospatial, with general overall cognition. This model suggested that there was adequate fit for the second-order model, with a CFI of 0.96, RMSEA of 0.03, and SRMR of 0.08 (Model 5, Table 2). The standardized loadings for the individual first-order factors with the higher-order cognition were 0.80 for visuospatial, 0.88 for language, 0.92 for orientation, and 0.98 for memory.
Criterion validity for age and education
The general cognition domain factor scores were significantly associated with age and education, with correlation coefficients of −0.28 (p < 0.01) and 0.57 (p < 0.01), respectively (Fig. 1).
A Correlation between general cognition factor scores and age. B Correlation between general cognition factor scores and education.
Discussion
In this study, we evaluated the psychometric properties of the Community Screening Interview for Dementia (CSI-D) participant interview using an a priori conceptualization of cognitive domains and found that the measure reasonably summarizes the hypothesized domains of orientation, language, memory, and visuospatial functioning in an aging population of urban Zambians. We found that, with thorough training and frequent reinforcement, CHWs could feasibly administer the CSI-D participant interview to community-dwelling participants across multiple local languages. CHWs appear well suited to improve access to dementia screening and management services in LMICs like Zambia. While the CSI-D participant interview offers an ideal starting point for a lay-administered assessment, the absence of items measuring executive functioning, a cognitive domain affected early in many disease processes that cause dementia, may limit its utility.
The CSI-D and the shorter, brief version of the CSI-D have been broadly used to screen for and assess dementia prevalence as part of research around the world, including in sub-Saharan Africa16,24. While components of the CSI-D participant interview are included in the Harmonized Cognitive Assessment Protocol (HCAP), which has been examined using confirmatory factor analysis in the United States and several HCAP partner studies including in sub-Saharan Africa31,32,33, the measurement structure of CSI-D participant interview has undergone comparatively little assessment of its factors in isolation. Prior studies have assessed test-retest reliability and inter-rater reliability of the CSI-D participant interview34,35 as well as compared its cutoff scores to other clinical dementia measures, largely in Asia30,36. Thus, this manuscript begins to address the critical question whether the CSI-D participant interview has sufficient psychometric validity to serve as a screen or diagnostic tool for dementia in LMICs like Zambia. In this factor analysis, we found that the CSI-D participant interview adequately fit four cognitive domains and that the general overall cognition factor score was positively correlated with educational attainment and inversely correlated with age. The domains of memory and language both were well-fitting models. For these domains, all included items were meaningfully related to the underlying domain as well as other items in the model. Memory items also showed adequate variability in item ability to assess varying levels of underlying ability in this cognitive domain. The delayed 3-world recall and CERAD immediate sum of three trials and CERAD word list delay are included in the Harmonized Cognitive Assessment Protocol, where performance on these measures has been variable across low-income settings where formal education is limited32,37. Further, while the modified CERAD has not been previously validated in Zambia, it is similar to the Hopkins Verbal Learning Test, which has locally validated norms38. Our findings are consistent with prior studies in which items in the memory domain show the greatest variability and offer the most discriminatory ability to assess cognition across diverse demographic groups32,37,39. Conversely, while model fit for language was satisfactory, item thresholds were less variable than in the memory domain (−1.99 to −0.71)32,37,40 suggesting that the CSI-D might have limitations in what can be assessed about this cognitive domain in our setting. While each of the orientation items had adequate factor loadings, item thresholds were uniformly low, except for the item “What is the name of the member of parliament.” This fits well with our observation that most Z-CHAD participants answered orientation items correctly and is also consistent with prior studies elsewhere that showed orientation items were most likely to exhibit a ceiling effect32. The lack of difficulty in these items makes them useful for diagnosing advanced dementia and screening for delirium, though they are likely less helpful for assessing smaller longitudinal changes in cognition33. For this reason, orientation to season and day of the week were intentionally included in the brief version of the CSI-D participant interview despite low item thresholds in order to screen for severe symptoms25.
The CSI-D participant interview and its brief counterpart do have notable shortcomings, namely the lack of items assessing executive functioning. Tests of executive functioning include tasks that require respondents to plan and utilize working memory, mental flexibility, and inhibitory control, among other skills. Executive functioning is often one of the earliest affected cognitive domains in diseases that cause dementia (e.g. vascular cognitive impairment), along with memory and some aspects of language, therefore their inclusion in cognitive testing in LMICs is essential in order to fully characterize the burden of mild cognitive impairment and mild dementia in these settings41,42. While other cognitive batteries include assessments of executive function, these tests often rely on a baseline level of numeracy and literacy and may not have cultural or sociocultural relevance across populations43. Another approach to assessing executive functioning is participant or informant reported impairment, which may have variable reliability based on cultural or recall biases44. Our study adds to the call for continued investment in culturally relevant assessments of executive function for low-education and low-literacy populations43,45,46.
A primary goal of this work was to determine whether CHWs could address a critical gap in dementia care in Zambia by administering a standardized cognitive tool to community members for the diagnosis of dementia. While prior studies suggest using the CSI-D as a screen for cognitive impairment followed by referral for more thorough assessment, this two-step approach is not feasible in Zambia due to a dearth of dementia experts, namely neurologists and neuropsychologists25. Therefore, there is a clear need for an accurate measure that can be used by non-specialists to screen for and facilitate dementia diagnosis. During this study, we found that lay healthcare workers who participated in a focused training on dementia, its assessment, and management, could feasibly administer the CSI-D participant interview and modified CERAD word recall. This is best demonstrated by the lack of missing data across two years of data collection with over 700 participants. Integral to our success was the inclusion of dementia caregivers in our team’s training. Their experiences improved our team’s understanding of dementia’s impact on Zambian patients and their families and offered motivation for this work. Regular field visits by our team’s neuropsychologists also helped address logistical barriers that arose when administering the CSI-D in the field, and biannual re-trainings ensured that all CHWs understood the utility of test items and their translation across multiple local languages.
This was a cross-sectional study in urban and peri-urban Lusaka communities. While most participants moved to Lusaka after completing their formal education, findings are likely not representative of cognitive function in other provinces, particularly rural regions. In addition, while CHWs were trained to administer the CSI-D participant interview in three languages, our prior clinical experiences suggested that participants would use multiple languages during data collection and as a result, we did not collect language of CSI-D participant interview item administration. A growing body of literature has shown that participants may perform differently when tested in their dominant versus a nondominant language47. As a result, some caution in interpreting verbal recall tests is warranted in multilingual settings like ours. Future studies in Zambia should consider assessment of measurement differences across languages. Informal interviewer translation of CSI-D participant interview items and responses could have introduced unmeasured measurement error or bias. While existing cognitive batteries in Zambia have largely been validated in English38, future studies in this setting should consider a more nuanced assessment of the impact of language on the psychometric properties of cognitive instruments. Lastly, as part of this study, data collection was conducted in four communities where there are differences in demographic characteristics known to influence cognitive test performance. There were significant differences in CSI-D participant interview item correct response rates when findings were stratified by study community. For this reason, future work should examine the possibility of differential item functioning in the CSI-D by factors previously shown to affect cognitive test performance in other settings48.
By 2050, 71% of people with Alzheimer’s Disease are expected to reside in LMICs like Zambia49. Improving access to dementia screening, diagnoses, and management will be critical to reduce the impact of dementia on patients, their families, and society. Our study found that CHWs can feasibly administer neurocognitive test items to address a critical diagnostic gap in the Zambian health sector, leveraging an established strength in health systems with CHWs. Yet, the CSI-D, likely one of the most widely utilized neurocognitive tools worldwide, does not fully capture crucial aspects of cognition in settings like ours. These findings suggest that the development of more nuanced, culturally relevant dementia measures is necessary to address the growing burden of dementia in LMICs.
Methods
Study population and design
The Zambian Cohort of Healthy Aging and Dementia (Z-CHAD) study is an ongoing observational study of dementia and associated risk factors in a population-based sample of urban, peri-urban, and rural Zambians. This analysis is drawn from participants recruited between August 7, 2023 and December 2, 2025 from four Lusaka communities - Chipata, Chawama, Chilenje, and Matero. These communities were selected because residents are a mix of urban and peri-urban poor, middle-, and upper-class and have a history of CHW involvement in community health promotion and education campaigns.
A multi-stage cluster sampling approach was used for this study with Zambian Census Agency enumeration areas (EAs) serving as the unit of sampling. As a population-based sample of older adults was not readily available and population proportions of EAs were not available from the 2022 National Census at the time of study initiation, we first selected a random sample of Zambian Census Agency enumeration areas in each study community using computerized random numbers. Enumeration areas (EAs) are the primary sampling unit for the national census50. In each EA, a team of CHWs conducted a door-to-door survey of households to enumerate residents. After enumeration, our teams returned to all households in the EA with eligible participants to discuss the study and obtained informed consent. In each household, a participant and informant were recruited. Eligible participants were 55 years or older and a resident of the household. As Zambia’s population remains relatively young, we selected this age cutoff to enable comparisons to prior studies of cognition in Zambia while also facilitating comparisons to other studies of aging and dementia in the region51. Informants were individuals aged 18 years or older who knew the participant well and would likely aid with activities of daily living or healthcare seeking, if necessary. Participants or informants who could not provide independent, informed consent in English or a local language (Nyanja or Bemba), were excluded.
Data were collected in the participant’s home and included demographics and socioeconomic characteristics, including years of formal education, place of birth, and household income. Place of birth and duration of residency in Lusaka were used to qualify education quality, which has varied between regions over time as education reforms have been implemented52,53. Medical history, including HIV-status was also assessed. Rapid HIV testing was conducted for consenting participants by CHWs certified in HIV counselling and testing54. Participant cognitive status was assessed using the 10/66 Dementia Protocol, which is comprised of the CSI-D participant interview and the modified Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) 10-word list learning task with delayed recall23,55. To limit the effect of participant fatigue, the CSI-D participant interview and modified CERAD were administered early once rapport was established with the interviewer. Though the participant was allowed to take breaks during the interview, the entire 50 min study visit was completed in one sitting. The CSI-D participant interview with modified CERAD required, on average, 30 min to administer.
Community-based cognitive assessment
Ten CHWS were trained to administer the CSI-D and 10-word modified CERAD list (Table 4)55. CHWs were residents of the communities in which they were recruiting and had all participated in community-based data collection for research. Each CHW was fluent in English as well as at least one additional local language (Nyanja or Bemba). To prepare for data collection, the CHWs underwent a ten-day training led by neurologists, neuropsychologists, and community caregivers of people with dementia. The training included lectures focused on dementia diagnosis and management, including best practices for cognitive testing in the home environment. Interactive sessions focused on common dementia symptoms, local terms and phrases used to describe these symptoms, and community beliefs. A Zambian neuropsychologist (LK) led instruction on the administration of the CSI-D participant interview and the 10-word modified CERAD list55. Prior to the training, the CSI-D was reviewed for appropriateness, as recommended by the WHO’s toolkit for preparing and adapting tools for implementation56. Two CSI-D participant interview items were modified to better fit the local context after consultation with neurologists and neuropsychologists with clinical experience in Zambia (LK, MM, ME). First, for phrase repetition, we used the Nyanja phrase “Kwina ku cita, kwina kuta cita”57,58,59. Second, instead of asking the name of a civil rights leader, participants were asked the name of the first Zambian president. During the training, the nuances of each item was discussed and translation to local languages and dialects were practiced until they were mastered. Role plays helped ensure that interview administration was accurate and consistent between CHWs. Instruments were provided to CHWs in English with the goal of administering the CSI-D participant interview in the language of the participant’s choice (English, Nyanja, Bemba) and the modified CERAD in English. We intentionally did not provide direct translations of the CSI-D participant interview or modified CERAD to local languages. While English is the official language and the primary language of formal education, Zambia has seven official languages and immense variation in dialects60. It is common for individuals to switch between languages during informal conversations and health care interviews61. Therefore, as our goal was to assess the utility of these tools in our real-world setting, there was concern that strict use of direct translations may inadvertently limit participant comprehension of test items and affect responses. For this reason, two neurologists (ME and MM) and a neuropsychologist (LK) spent extensive time ensuring study staff understood each question and agreed on common terms across languages for each question to ensure conceptual equivalence between languages. After training, CHWs pre-tested the CSI-D participant interview and modified CERAD with community members with no cognitive symptoms, as well as members of the Alzheimer’s Association of Zambia (ADDIZ; local Alzheimer’s International affiliate) with known dementia diagnoses. We then conducted cognitive debriefing, to identify confusing phrasing, with the CHWs and pre-test participants and no additional adaptions were made.
Data collection was conducted in or outside the participant’s home by a gender-concordant interviewer with fluency in the participant’s preferred language. Possible distractions were minimized whenever possible. While the informant was allowed to aid with some of the participant’s interview questions, namely those pertaining to medical history, they were separated from the participant during administration of the CSI-D participant interview and modified CERAD to avoid aiding the participant. As in the 10/66 Dementia Protocol, the modified CERAD 10-item word list was administered as three separate trials55. After each trial, participants were asked to recall as many items as possible. After a five-minute delay, the participant was again asked to recall the items. To ensure accuracy of test administration, one neuropsychologist and two graduate trainees in neuropsychology would routinely observe CSI-D participant interview and modified CERAD test administration in the field by each CHW and provide feedback when necessary. Every six months, the team held in-person debriefings on data collection to identify discrepancies and to retrain staff in test administration.
Data management and analysis
Data were entered into REDCap in the field, except for visuospatial items, which were completed on paper and scored electronically and verified for accuracy before analysis62. Participant demographics were summarised by counts (%) for categorical variables and mean (SD) for continuous variables. The median (Q1, Q3) was used in addition to the mean (SD) for potentially skewed variables. To assess the feasibility of administering the CSI-D participant interview across populations with different demographic characteristics known to impact cognitive test performance, participants were grouped by study community and compared using chi-square, one-way analysis of variance (ANOVA), or the appropriate non-parametric test.
Participant responses to each item were scored as correct or incorrect, and individual test items were handled as shown in Table 421,55. We then grouped the items based on underlying cognitive domains. A team including two neuropsychologists (LK and EB) classified each item using categories proposed by the original developers of the CSI-D, the Indianapolis-Ibadan Dementia Project, theory, and prior empirical work (Table 4)37,63.
Confirmatory factor analysis (CFA) was used to assess the measurement structure of the CSI-D participant interview in Zambia. CFA attempts to explain relationships between unobservable characteristics, known as latent factors, and observable outcomes64. In this study, latent factors refer to cognitive domains that are measured through CSI-D responses by participants. CFA models allow us to estimate how well our defined model fits the observed data to confirm if the CSI-D participant interview items align with the intended factors. To evaluate the relationship between CSI-D participant interview items and underlying cognitive domains, we performed a series of confirmatory factor analyses (CFAs). We fit a series of unidimensional, single-factor models for each cognitive domain and its corresponding CSI-D participant interview items to ensure items fit into their theoretical cognitive domain. Then, we fit a second-order factor model to assess the relationship between the first-order cognitive domains and a higher-order domain, such as overall general cognition. Due to the presence of both continuous and categorical variables in the CSI-D participant interview, we used the Diagonally Weighted Least Squares (DWLS) estimator for our CFA models. Models were fit by freely estimating variance and loadings except for the visuospatial factor, which only had two items and was thus fit with fixed loadings.
To assess model fit for each CFA, we used the Comparative Fit Index (CFI) the Root Mean Square Error of Approximation (RMSEA), and the Standardised Root Mean Square Residual (SRMR). Common thresholds for well-fitting models include CFI ≥ 0.95, RMSEA < 0.06, and SRMR ≤ 0.08, while acceptable goodness of fit include CFI ≥ 0.90, RMSEA ≤ 0.08, and SRMR ≤ 0.0865,66. To assess monotonicity and item correlation with other items within the cognitive domain, standardized loadings were calculated in each CFA model. A threshold for standardized loadings greater than or equal to 0.40 was considered acceptable67. To assess the level of each cognitive domain required for participants to answer an item correctly, we reported thresholds for categorical items and intercepts for continuous items. If model fit indices indicated poor fit, we then assessed individual items that may have contributed to model misfit, such as low standardized loadings, high correlation with other items, or low variability in responses, and then removed individual items to evaluate improvement in model diagnostics. Residual correlations were included between items that violated the assumption of local independence.
Finally, to assess construct validity, we evaluated the associations between factor scores and participant age and education for the general cognition domain. Specifically, Spearman’s correlation coefficients were used to assess the relationship between age and education with the factor score for the general cognition domain separately.
Statistical significance for participant comparisons between study communities was assessed using a p-value threshold of 0.05. All analyses, including the CFAs, were conducted in R version 4.5.0. As the goal of this analysis was to assess whether the CSI-D participant interview appropriately captures the cognitive domains that they were designed to test in Zambia and not to assess dementia prevalence in this population, we did not calculate the CSI-D sum scores (cogscore, relscore or dfscore) and CSI-D informant data is not included in this analysis.
Standard protocol approvals, registrations, and consents
Written, informed consent was obtained from all participants in English, Nyanja, or Bemba. Those that could not sign their name placed a thumbprint on the consent form. Ethical approval was obtained from University of Michigan Institutional Review Boards and the University of Zambia Biomedical Research Ethics Committee, with final approval from the Zambian National Health Research Authority.
Data availability
Data generated and analyzed during this current study are not publicly available as the study is still ongoing but de-identified data are available from the corresponding author upon reasonable request and a signed data sharing agreement.
References
Collaborators, G. B. D. D Global, regional, and national burden of Alzheimer’s disease and other dementias, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol. 18, 88–106, https://doi.org/10.1016/S1474-4422(18)30403-4 (2019).
Prince, M., Guerchet, M. & Prina, M. The Global Impact of Dementia 2013-2050. (Alzheimer’s Disease International, London, 2013).
Zambia Statistics Agency. Zambia Data Portal, 2022).
Republic of Zambia Ministry of Health. (ed Mother and Child Health Ministry of Community Development) (2015).
Sharma, A. et al. The Social Construction of Aging Among a Clinic-Based Population and Their Healthcare Workers in Zambia. Int. J. Public Health 69, https://doi.org/10.3389/ijph.2024.1606607 (2024).
World Health Organization. ATLAS Country Resources for Neurological Disorders. Vol. 2nd (2017).
Zambia Statistics Agency. Summary Report. 148 (Zambia Statistics Agency, Lusaka, 2022).
Yoseph, M. et al. Prevalence estimates of dementia in older adults in rural Kilimanjaro 2009-2010 and 2018-2019: is there evidence of changing prevalence? Int J. Geriatr. Psychiatry 36, 950–959, https://doi.org/10.1002/gps.5498 (2021).
George, G. et al. Screening for delirium and dementia in older hospitalised adults in Zambia. J. Neurol. Sci. 436, 120186. https://doi.org/10.1016/j.jns.2022.120186 (2022).
Jacobs, R., Schneider, M., Farina, N., du Toit, P. & Evans-Lacko, S. Stigma and its implications for dementia in South Africa: a multi-stakeholder exploratory study. Ageing Soc. 1–31, https://doi.org/10.1017/s0144686x2200040x (2022).
Kehoua, G. et al. People with Dementia in Sub-Saharan Africa: From Support to Abuse by Caregivers: Results of EPIDEMCA-FU Program in Congo. Dement Geriatr. Cogn. Dis. Extra 9, 163–175, https://doi.org/10.1159/000489846 (2019).
World Health Organization. What do we know about community health workers? A systematic review of existing reviews. (World Health Organization Health Workforce, 2021).
Mwai, G. W. et al. Role and outcomes of community health workers in HIV care in sub-Saharan Africa: a systematic review. J. Int AIDS Soc. 16, 18586. https://doi.org/10.7448/IAS.16.1.1858618586 (2013).
Stoutenberg, M. et al. Acceptability and feasibility of home-based hypertension and physical activity screening by community health workers in an under-resourced community in South Africa. Z. Gesundh Wiss. 1–12 https://doi.org/10.1007/s10389-023-01873-w (2023).
Paddick, S. M. et al. Effectiveness of app-based cognitive screening for dementia by lay health workers in low resource settings. a validation and feasibility study in rural Tanzania. J. Geriatr. Psychiatry Neurol. 34, 613–621, https://doi.org/10.1177/0891988720957105 (2021).
de Jager, C. A., Msemburi, W., Pepper, K. & Combrinck, M. I. Dementia prevalence in a rural region of South Africa: a cross-sectional community study. J. Alzheimers Dis. 60, 1087–1096, https://doi.org/10.3233/JAD-170325 (2017).
Hoops, S. et al. Validity of the MoCA and MMSE in the detection of MCI and dementia in Parkinson disease. Neurology 73, 1738–1745, https://doi.org/10.1212/WNL.0b013e3181c34b47 (2009).
Lwere, K. et al. Cognitive screening and education: assessing the Montreal Cognitive Assessment’s validity in older Ugandan populations. Acad. Ment. Health Well Being 2 https://doi.org/10.20935/mhealthwellb7804 (2025).
Pellicer-Espinosa, Isabel. A. U. D. I.- O. Cognitive screening instruments for older adults with low educational and literacy levels: a systematic review. J. Appl. Gerontol. 41, 1222–1231 (2022).
Wu, Y., Zhang, Y., Yuan, X., Guo, J. & Gao, X. Influence of education level on MMSE and MoCA scores of elderly inpatients. Appl. Neuropsychol.: Adult. 30, 414–418 (2025).
Hall, K. et al. The development of a dementia screening interview in two distinct languages. Int. J. Methods Psychiatr. Res. 3, 1–28 (1993).
Hendrie, H. C. et al. Incidence of dementia and Alzheimer disease in 2 communities: Yoruba residing in Ibadan, Nigeria, and African Americans residing in Indianapolis, Indiana. JAMA 285, 739–747, https://doi.org/10.1001/jama.285.6.739 (2001).
Hall, K. S. et al. Community screening interview for dementia (CSI ‘D’); performance in five disparate study sites. Int. J. Geriatr. Psychiatry 15, 521–531, 10.1002/1099-1166(200006)15:6<521::aid-gps182>3.0.co;2-f (2000).
Prince, M. et al. Dementia incidence and mortality in middle-income countries, and associations with indicators of cognitive reserve: a 10/66 Dementia Research Group population-based cohort study. Lancet 380, 50–58, https://doi.org/10.1016/S0140-6736(12)60399-7 (2012).
Prince, M. et al. A brief dementia screener suitable for use by non-specialists in resource poor settings–the cross-cultural derivation and validation of the brief Community Screening Instrument for Dementia. Int J. Geriatr. Psychiatry 26, 899–907, https://doi.org/10.1002/gps.2622 (2011).
de Jager, C. A. & Pepper, K. Building capacity for dementia screening and intervention in low income, rural communities: A collaborative initiative. Int. J. Afr. Nurs. Sci. 7, 43–49, https://doi.org/10.1016/j.ijans.2017.07.001 (2017).
Musyimi, C. W., Ndetei, D. M., Muyela, L. A., Masila, J. & Farina, N. Dementia Screening in Rural Kenya: The Prevalence and Impact of Screening Positive for Dementia. Neuroepidemiology 58, 174–181, https://doi.org/10.1159/000536012 (2024).
Farina, N. et al. Banerjee & on behalf of the STRiDE team. A systematic review and meta-analysis of dementia prevalence in seven developing countries: A STRiDE project. Glob. Public Health 15, 1878–1893 (2020).
Crane, P. K. et al. Item response theory facilitated co calibrating cognitive tests and reduced bias in estimated rates of decline. J. Clin. Epidemiol. 61, 1018–1027.e1019, https://doi.org/10.1016/j.jclinepi.2007.11.011 (2008).
Venketasubramanian, N. A Comparative Study of Three Dementia Screening Instruments (CSI-D, CMMSE, and ECAQ) in a Multi-Ethnic Asian Population. Healthcare (Basel) 12 https://doi.org/10.3390/healthcare12030410 (2024).
Jones, R. N. et al. Factor structure of the Harmonized Cognitive Assessment Protocol neuropsychological battery in the Health and Retirement Study. J. Int Neuropsychol. Soc. 30, 47–55, https://doi.org/10.1017/S135561772300019X (2024).
Gross, A. L. et al. Harmonisation of later-life cognitive function across national contexts: results from the Harmonized Cognitive Assessment Protocols. Lancet Healthy Longev. 4, e573–e583, https://doi.org/10.1016/S2666-7568(23)00170-8 (2023).
Gross, A. L. et al. Midlife cognitive testing in Africa: validity of the Harmonised Cognitive Assessment Protocol in the Kenya Life Panel Survey. BMJ Open 15, e096619, https://doi.org/10.1136/bmjopen-2024-096619 (2025).
Hall, K. et al. A cross-cultural community based study of dementias: methods and performance of the survey instrument, Indianapolis, U.S.A., and Ibadan, Nigeria. Int. J. Methods Psychiatr. Res. 6, 129–142 (1996).
Liu, S. I. et al. Validity and reliability of a Taiwan Chinese version of the community screening instrument for dementia. Am. J. Geriatr. Psychiatry 13, 581–588, https://doi.org/10.1176/appi.ajgp.13.7.581 (2005).
Chan, T. S., Lam, L. C., Chiu, H. F. & Prince, M. Validity and applicability of the Chinese version of community screening instrument for dementia. Dement Geriatr. Cogn. Disord. 15, 10–18, https://doi.org/10.1159/000066672 (2003).
Nichols, E. et al. Differences in the measurement of cognition for the assessment of dementia across geographic contexts: recommendations for cross-national research. Alzheimers Dement 19, 1009–1019, https://doi.org/10.1002/alz.12740 (2023).
Hestad, K. A. et al. Do neuropsychological test norms from African Americans in the United States generalize to a Zambian population? Psychol. Assess. 28, 18–38, https://doi.org/10.1037/pas0000147 (2016).
McGrory, S., Doherty, J. M., Austin, E. J., Starr, J. M. & Shenkin, S. D. Item response theory analysis of cognitive tests in people with dementia: a systematic review. BMC Psychiatry 14, 47. https://doi.org/10.1186/1471-244X-14-47 (2014).
Serpell, R. How the study of cognitive growth can benefit from a cultural lens. Perspect. Psychol. Sci. 12, 889–899, https://doi.org/10.1177/1745691617704419 (2017).
Jutten, R. J. et al. Identifying sensitive measures of cognitive decline at different clinical stages of Alzheimer’s Disease. J. Int Neuropsychol. Soc. 27, 426–438, https://doi.org/10.1017/s1355617720000934 (2021).
Junquera, A., García-Zamora, E., Olazarán, J., Parra, M. A. & Fernández-Guinea, S. Role of executive functions in the conversion from mild cognitive impairment to dementia. J. Alzheimer’s. Dis. 77, 641–653, https://doi.org/10.3233/jad-200586 (2020).
Laurence, P. G. Executive function beyond minority world contexts: a call for culturally relevant assessments in historically overlooked populations in majority world countries. Mind, Brain, Educ. 20, e70034. https://doi.org/10.1111/mbe.70034 (2026).
Baggetta, P. & Alexander, P. Conceptualization and operationalization of executive function. Mind, Brain, Educ. 10, 10–33 (2016).
Kobayashi, L. C., Wardle, J., Wolf, M. S. & von Wagner, C. Cognitive Function and Health Literacy Decline in a Cohort of Aging English Adults. J. Gen. Intern Med 30, 958–964, https://doi.org/10.1007/s11606-015-3206-9 (2015).
Kosmidid, M. H. Challenges in the neuropsychological assessment of illterate older adults. Lang., Cogn. Neurosci 33, 373–386 (2018).
Gavino, M. F., Salmon, D. P., Mizrahi, R. & Gollan, T. H. Language dominance effects on verbal list memory in older Spanish-English bilinguals. Neuropsychology 40, 282–295, https://doi.org/10.1037/neu0001058 (2026).
Goel, A. & Gross, A. Differential item functioning in the cognitive screener used in the Longitudinal Aging Study in India. Int Psychogeriatr. 31, 1331–1341, https://doi.org/10.1017/S1041610218001746 (2019).
Alzheimer’s Disease International. World Alzheimer Report 2015: The global impact of dementia. An analysis of prevalence, incidence, costs and trends. (London, 2015).
Zambia Statistics Agency. 2022 Living Conditions Monitoring Survey Report. (Zambia Statistics Agency, Lusaka, 2024).
Farrell, M. T. et al. Estimating dementia prevalence using remote diagnoses and algorithmic modelling: a population-based study of a rural region in South Africa. Lancet Glob. Health 12, 2003–2011, https://doi.org/10.1016/S2214-109X(24)00325-5 (2024).
United Nations Educational, S. A. C. O. Spotlight on basic education completion and foundational learning: Zambia. 44 (France, 2024).
Siakalima, S. & Kanyamuna, V. Community Schooling System in Zambia: Its Evolution and Stakeholder Perspectives. Am. J. Educ. Res. 10, 571–578 (2022).
Ministry of Health. Zambia Consolidated Guidelines for Treatment and Prevention of HIV Infection. (Ministry of Health, 2022).
Prince, M. et al. The protocols for the 10/66 dementia research group population-based research programme. BMC Public Health 7, 165. https://doi.org/10.1186/1471-2458-7-165 (2007).
World Health Organization. Preparing and adapting tools for implementation, 2026).
Holguin, A. et al. HIV-1 effects on neuropsychological performance in a resource-limited country, Zambia. AIDS Behav. 15, 1895–1901, https://doi.org/10.1007/s10461-011-9988-9 (2011).
Kvalsund, M. P., Haworth, A., Murman, D. L., Velie, E. & Birbeck, G. L. Closing gaps in antiretroviral therapy access: human immunodeficiency virus-associated dementia screening instruments for non-physician healthcare workers. Am. J. Trop. Med Hyg. 80, 1054–1059 (2009).
Kalungwana, L. et al. Cognitive impairment and psychiatric morbidity in HIV+ Zambians with new-onset seizure. Am. J. Trop. Med Hyg. 91, 1254–1258, https://doi.org/10.4269/ajtmh.13-0758 (2014).
Banda, F. & Jimaima, H. Linguistic landscapes and the sociolinguistics of language vitality in multilingual contexts of Zambia. Multilingua 36, 595–625, https://doi.org/10.1515/multi-2017-3047 (2017).
Simngala, G. & Jimaima, H. Multilingual realitives of lanugage contact at the University of Zambia. J. Asian Afr. Stud. 56, 1644–1657 (2021).
Harris, P. A. et al. The REDCap consortium: building an international community of software platform partners. J. Biomed. Inf. 95, 103208. https://doi.org/10.1016/j.jbi.2019.103208 (2019).
Project, I.-I. D. CSID Domain Scores, 2021).
Embretson, S. & Preise, S. Item response theory for psychologists. (Lawrence Erlbaum Associates, Inc., 2000).
Hooper, D., Coughlan, J. & Mullen, M. Structural equation modelling: guidelines for determining model fit. EJBRM 6, 53–60 (2008).
Hu, L. & Bentler, P. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct. Equ. Modeling: A Multidiscplinary J. 6, 1–55, https://doi.org/10.1080/10705519909540118 (1999).
Pituch, K. & Stevens, J. (Taylor & Francis Group, Abingdon, Oxon, 2016).
Acknowledgements
This study is funded by the University of Michigan’s Center to Accelerate Population Research in Alzheimer’s (NIH NIA P30AG066582), the Center for Global Health Equity Data Collaborative, the African Studies Center, and Global REACH. The funders played no role in study design, data collection, analysis, and interpretation of data, or the writing of this manuscript. Dr. Elafros is supported by the NIH NINDS (K23NS131444) and the Andrea and Lawrence A. Wolfe Research Professorship. Dr. Tanner is supported by the NIH NIA (K08AG090815).
Author information
Authors and Affiliations
Ciheb Zambia, Lusaka, Zambia
Lisa Kalungwana, Lottie M. Hachaambwa, Linah Mwango & Melissa A. Elafros
University of Zambia, Department of Psychiatry, Lusaka, Zambia
Lisa Kalungwana
University of Michigan, Department of Neurology, Ann Arbor, MI, USA
Nicholas Heaton & Melissa A. Elafros
University of Michigan, Department of Physical Medicine and Rehabilitation, Ann Arbor, MI, USA
Emily Briceno
Ministry of Health, Chipata Level One Hospital, Lusaka, Zambia
Mataa M. Mataa
Division of Infectious Diseases, Department of Medicine, University of Maryland School of Medicine, Baltimore, MD, USA
Lottie M. Hachaambwa
Glenn Biggs Institute for Alzheimer’s and Neurodegenerative Diseases, Department of Neurology, University of Texas Health Science Center San Antonio, San Antonio, TX, USA
Jeremy A. Tanner
Authors
- Lisa Kalungwana
- Nicholas Heaton
- Emily Briceno
- Mataa M. Mataa
- Lottie M. Hachaambwa
- Linah Mwango
- Jeremy A. Tanner
- Melissa A. Elafros
Contributions
L.K. and M.E. were involved in the study design, data collection, interpretation of the statistical analysis, and wrote the manuscript. N.H. performed the statistical analysis, interpreted the analysis and wrote the manuscript. E.B. was involved in the interpretation of results and providing critical revisions to the manuscript. M.M. was involved in the study design, data collection and providing critical revisions to the manuscript. L.M. was involved in data collection and provided critical revisions to the manuscript. J.T. and L.H. were involved in the study design and providing critical revision to the manuscript.
Corresponding author
Correspondence to Melissa A. Elafros.
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Kalungwana, L., Heaton, N., Briceno, E. et al. Factor structure of the Community Screening Interview for Dementia (CSI-D) among Zambian adults. npj Dement. 2, 99 (2026). https://doi.org/10.1038/s44400-026-00142-3
Received: 20 April 2026
Accepted: 25 August 2026
Published: 27 October 2026
Version of record: 27 October 2026
DOI: https://doi.org/10.1038/s44400-026-00142-3
来源:Nature 健康护理主题 · nature.com