It is by now known that the factors that contribute to frailty in SSA may differ from HICs due to regional variations in the demographic, social and economic landscapes34. Screening for frailty in this region, therefore, requires further exploration, especially regarding the instruments and thresholds used for identifying older people at risk for frailty. Consequently, this study aims to develop a regional specific frailty threshold for older people in SSA using pooled data from multiple countries. The findings of our study indicate that frailty thresholds vary across SSA countries, with country-specific thresholds ranging between 0.24 and 0.32. A pooled threshold of 0.29 was established through meta-analysis, accounting for heterogeneity across populations. Developing a frailty threshold for older people in SSA can help in the early detection of frailty with improved accuracy and relevance, as well as enhance effective allocation of resources to improve the well-being of older people.
We employed a cross-sectional design using multiple datasets from four countries in SSA (Kenya, Ghana, Uganda and Côte d’Ivoire). The datasets were sourced from various population-based studies focusing on older people (50 years and above) in SSA. We used data from the Health and Well-being of Older Persons study in Kenya (HWOPs-1), Wave 2 of the WHO Study on Global AGEing and adult health (SAGE) Ghana, the WHO SAGE-WOPS HIV sub-study in Uganda, and the Living Condition, Health and Resilience among the Elderly study in Côte d’Ivoire. Overall, 783 older people (60 years and above) were included from Kenya, 3266 older people (50 years and above) from Ghana, 461 (50 years and above) from Uganda and 1017 (50 years and above) from Côte d’Ivoire. Demographic information (age and sex) and health information were obtained from the respective datasets for this study. Access to the datasets was granted upon formal application to the respective data custodians. Each request included a description of the study objectives and data use agreements. This study was approved by the Torrens University Australia Human Research Ethics Committee (Approval number: 0353).
This study was designed to develop a research framework for routine generation of evidence on the health and well-being of older people (60 years and above) in Kenya and developing and piloting an essential research tool with key indicators to enable rapid and routine assessment of the health of older people. The specific objectives of the study were to develop and validate a research tool to assess the health and well-being of older people in Kenya, examine the disease and disability burden among older people in a selected county, identify health and socio-economic concerns and needs that affect the well-being of the older persons, identify strategies that will enhance the health, psycho-social and general well-being of older people, and strengthen the research capacity of the collaborating institutions through designing a policy-focused study and production of research. The study was conducted in Kiambu County in the central region of Kenya. Kiambu is one of the counties bordering Nairobi, the capital city of Kenya. A cross-sectional survey design was used, where households were selected through a multi-stage random sampling of households with older persons. The first stage involved a random selection of 30 clusters in the National Sample Survey and Evaluation Programme (NASSEP). The second stage involved the identification of households with older persons in each of the selected clusters, and 10 households were randomly selected per cluster. About 300 households were selected using random sampling with replacement to account for non-response. Data were collected electronically using tablets and uploaded daily to central servers which were monitored for completeness and quality. The dataset is owned and hosted by the African Population and Health Research Centre, Nairobi, Kenya. HWOPs-1 was approved by the Kenyatta University Ethical Review Committee and the Scientific Steering Committee (Ref. No. PKU/8691934). Oral informed consent was obtained from the participants.
The WHO Study on Global AGEing and adult health (SAGE) is a nationally representative survey conducted in Ghana through multistage cluster sampling strategies. The survey is a multi-country longitudinal study that collects data to complement existing ageing data sources to inform policy and programmes. WHO and the University of Ghana Medical School through Department of Community Health collaborated to implement SAGE Wave 2 in 2014-2015. Individuals aged ≥50 years were interviewed regarding their chronic health conditions and health services coverage; subjective well-being and quality of life; health care utilization; risk factors and preventive health behaviours; perceived health status; socio-demographic and work history; social cohesion; and household characteristics. Similar information was collected on smaller sample of persons aged 18-49 years. In households identified as “older” for sampling purposes, all household members aged 50 years and older were invited to participate in the study. SAGE was approved by the World Health Organization’s Ethical Review Board (reference number RPC149) and the Ethical and Protocol Review Committee, College of Health Sciences, University of Ghana, Accra, Ghana. Written informed consent was obtained from all study participants.
The Well-Being of Older People Study is the second round of the survey (WOPS)-2013. WOPS surveys are designed by the WHO and the Medical Research Council of Uganda and implemented by the Medical Research Council of Uganda. The objectives of the data collection were to describe the roles, health problems (physical and mental) and social well-being of older people who are directly and indirectly affected by HIV/AIDS, with special attention to the effects of the introduction of Anti-Retroviral Therapy (ART), and to develop recommendations for policy and practice that can be expected to improve the well-being of older people affected by or infected with HIV/AIDS. Individuals aged ≥50 years were interviewed regarding respondent and household characteristics, health state description, chronic conditions and health service coverage, health care utilization and risk factors and behaviour, health measurements, care giving and care receiving roles. SAGE-WOPS HIV sub-study was approved by Uganda Virus Research Institute Research and Ethics Committee, the Uganda National Council for Science and Technology and the WHO Ethical Review Committee (RPC-149). All study respondents gave a written/thumb printed consent to participate in the study.
The dataset was obtained by combining data from three surveys in 3 regions of Côte d’Ivoire. The first survey was conducted in the department (sub-region) of Toumodi, Central Côte d’Ivoire, in July 2018. The second survey was conducted in the sub-prefecture of Daloa, western Côte d’Ivoire in November 2023, while the third survey was conducted in the sub-prefecture of Korhogo, Northern Côte d’Ivoire in May 2024. A two-stage sampling strategy was used. At the first stage, 51 enumeration areas for Toumodi, 22 enumeration areas for Daloa, and 29 enumeration areas for Korhogo were randomly selected. At the second stage, 30 households were randomly selected in each enumeration area. In each selected household, all individuals aged 50 years and above were preselected as participants in the aging questionnaire. Participants in the survey were definitely included upon oral consent. The final sample sizes were 557 in Toumodi, 278 in Daloa, and 242 in Korhogo. These surveys used the same questionnaire. Information on both perceived and observed health conditions and health behaviour, perceived survival, socioeconomic and demographic characteristics of participants was collected. Face to face interviews were conducted by ENSEA students in French under the supervision of their assistant professors. On average, each interview took approximately 35 min to complete. Data were collected using tablets loaded with CSPro software and subsequently imported into Stata v.17.0. The Living Condition, Health and Resilience among the Elderly study was approved by the National Ethics Committee of Côte d’Ivoire (Comité Consultatif National de Bioéthique de la République de Côte d’Ivoire).
The FI was adopted for frailty assessment in this study. The FI is a validated tool adopted globally to assess frailty among older people across community, acute and subacute settings. The FI is based on the accumulation of health deficits, which include a range of physical, social and cognitive indicators. Each deficit contributes equally to the overall score, and the index is calculated as the ratio of the number of deficits present to the total number of deficits considered.
In this study, the FI was developed from the four datasets following the recommendations by Rockwood and colleagues. Deficits included in our FI covered a range of health domains (physical health, functional ability, mental health, sensory function and social well-being). We developed our FI considering the contextual factors (culture, healthcare access, social integration, etc.) in the SSA region. We also considered the limited healthcare access in most parts of SSA, and accordingly limited the number of chronic diseases included in our FI. The FI items were discussed among the team of authors who are familiar with aging and healthcare in SSA.
We identified 30 items in each dataset that met the standard technical criteria. Each of the FI items was scored such that 0=deficit absent and 1 = deficit present. The scores were added and divided by the number of items (30) to create a variable between 0.00 (no deficits present) and 1.00 (all deficits present). The FI items were largely consistent across the four datasets, ensuring a high level of consistency in the variables used for the analysis. The internal consistency (Cronbach’s α values) was high across all datasets, ranging from 0.89 (Côte d’Ivoire) to 0.94 (Ghana), indicating good reliability of the FI. Details of the respective FI items across the datasets are presented in the Supplementary Information (see Supplementary Tables 1-4).
A 6-item outcome variable assessing dependency and independence in Activities of Daily Living (ADLs) was derived from the datasets from Kenya (Cronbach’s α: 0.85), Ghana (Cronbach’s α: 0.84), Uganda (Cronbach’s α: 0.74), and Côte d’Ivoire (Cronbach’s α: 0.71) indicating acceptable internal consistency of the items. We coded ADL performance as a binary variable, where participants who reported independence in all six ADLs were coded as 1 (independent), and those reporting dependence in one or more ADLs were coded as 0 (dependent). Frailty is a strong predictor dependency.
Receiver Operating Characteristic (ROC) analysis was performed on each dataset to determine the optimal thresholds. The FI was used as test variables, and the binary outcome measures (dependency versus independence with ADL) were used as state variables. The Area Under the ROC Curve (AUC) was calculated to assess the overall performance of the FI in discriminating frailty status in each of the datasets. The AUC summarizes the overall diagnostic accuracy of a classification test, with a value of 0.50 suggesting no discrimination, 0.50 to 0.70 being poor discrimination, 0.70-0.80 suggesting good discrimination, 0.80-0.90 suggesting very good discrimination, and 0.90-1.00 suggesting excellent discriminatory power. The optimal thresholds were determined using the AUC, and J statistics (Youden Index) which maximizes the sum of sensitivity and specificity. This process involved plotting the true positive rate (sensitivity) against the false positive rate (1-specificity) at various threshold levels. This process helped to prevent the influence of disease prevalence on predictive values often seen in Positive and Negative Predictive Values (PPV and NPV). The ROC analysis was conducted under non-parametric assumption using IBM SPSS version 29.20. Statistical significance for all analyses was set at P value < 0.05.
After determining thresholds from each dataset, we explored the possibility of combining these thresholds into a single, unified threshold to overcome the spectrum effect (variation in diagnostic test performance across different populations and subgroups due to disease prevalence, severity of condition, and other population characteristics) often associated with single population ROC based thresholds. This involved using a random-effects (Restricted Maximum Likelihood) meta-analysis technique to aggregate the ROC results from the different datasets. Heterogeneity was assessed using tau-squared (τ²), the I² statistic, and Cochran's Q test. By pooling the AUC, sensitivity, specificity, standard error and threshold values from each dataset, we derived a unified threshold. Meta-analysis of diagnostic accuracy tests has been recommended and utilized in several epidemiological and clinical literature. Subgroup analyses were conducted to explore potential sources of heterogeneity in our frailty threshold estimates. First, countries were grouped into two regional categories: West Africa (Ghana, Côte d'Ivoire) and East Africa (Kenya, Uganda) to assess geographical differences. Second, the frailty thresholds derived from the ROC analyses were used to classify countries into high threshold (above the pooled threshold) and low threshold (below the pooled threshold) groups, allowing for comparisons based on frailty burden. The meta-analyses were conducted using STATA version 16.0.
To evaluate the practical utility of the pooled threshold, we applied it to each country-specific dataset and calculated the corresponding sensitivity and specificity values. This step provided a secondary validation of the pooled threshold's discriminative performance across the diverse national contexts.
We conducted a meta-regression using a random-effects model with Restricted Maximum Likelihood (REML) estimation to explore study-level factors contributing to the observed heterogeneity in the pooled threshold. The moderators examined included mean-centred values for age and percentage of female participants. We tested each moderator in a separate model to assess their independent association with the reported threshold. The Knapp-Hartung adjustment was applied to improve the accuracy of confidence intervals given the small number of studies. Between-study variance explained by each model was quantified using the R² statistic. Residual heterogeneity was assessed using tau-squared (τ²), the I² statistic, and Cochran's Q test.
All statistical analyses were conducted using IBM SPSS version 29.20 and STATA version 16.0. ROC analyses were performed independently for each country dataset to determine optimal FI thresholds using Youden's Index (sensitivity + specificity - 1), with dependency in ADLs as the binary outcome. For each model, AUC, sensitivity, specificity, standard errors, and 95% confidence intervals were calculated under non-parametric assumptions.
To obtain a unified threshold, a random-effects meta-analysis using REML estimation was conducted. Statistical heterogeneity was assessed using I², τ², and Cochran's Q test. Statistical significance was set at p < 0.05 for all analyses. Subgroup analyses were performed by region (West vs. East Africa) and frailty burden (low vs. high thresholds). Meta-regression was conducted to examine study-level moderators (mean age and percentage of female participants), with the Knapp-Hartung adjustment applied to improve inference precision. Model performance was quantified using R², and residual heterogeneity was reported.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
