Temporal patterns of haemoglobin phenotypes among Nigerian university students based on ten years of surveillance data Authors: Thomas Olumide Adeleke, Chukwudumebi Placid Ughonu, Efeturi Agelebe, Oludamola Victoria Adeleke, Taiwo Omotayo Dosumu, Dolapo Emmanuel Ajala, Bayonle Ayenigba, Samuel Ademola Adegoke, Tina M. Slusher, Alexander A. Boucher Published: 2026 ABSTRACT Haemoglobinopathies remain among the most common monogenic disorders worldwide, with sickle cell disease (SCD) and various forms of thalassemia posing significant challenges to public health systems, particularly in low- and middle-income countries. This study presents a ten-year retrospective analysis of haemoglobin phenotype surveillance data, encompassing multiple regions and demographic strata. This retrospective work was conducted at Bowen University Hospital, Iwo, Osun State, southwestern Nigeria between January 2015 and December 2024. Data were extracted and analyzed using machine learning from 8890 medical records of undergraduate annual phenotype screening spanning ten years (2015–2024). HbAA was the most common genotype at 75.11% (n = 6,677). SCD genotypes (HbSS and HbSC) showed a non-significant downward trend over the ten-year period (slope = − 0.03 percentage points/year; R2 = 0.07; p = 0.45), and their combined prevalence remained stable at approximately 1.2%. Carrier prevalence (HbAS and HbAC) also remained stable across the study period. SCD continues to represent an important public health concern, particularly in resource-limited settings, where it contributes to morbidity, mortality, and reduced educational participation. This ten-year dataset, drawn from a single university cohort, offers a useful vantage point from which to reflect on the direction of institutional haemoglobinopathy surveillance. Because the findings are derived from admitted university students, they should be interpreted as indicative of a relatively healthy, educated young adult population rather than of Nigeria’s general population. The approach may nonetheless inform and support the design of broader institutional surveillance, policy engagement, and community-level health-education efforts.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-59392-z. Introduction Haemoglobinopathies remain among the most common monogenic disorders worldwide, with sickle cell disease (SCD) and various forms of thalassemia posing significant challenges to public health systems, particularly in low- and middle-income countries1. The global burden of haemoglobinopathies is closely tied to the distribution of haemoglobin (Hb) variants, especially in regions where malaria is or was endemic2. In routine clinical and epidemiological screening using cellulose acetate electrophoresis, these haemoglobin variants are identified at the protein level as phenotypes, which are conventionally reported as Hb genotypes (e.g., HbAA, HbAS, HbSS). Understanding the epidemiology of these haemoglobin variants over time provides critical insights into both inherited disease patterns and evolving public health responses. The importance of long-term Hb phenotype (commonly reported as genotype) surveillance lies in its ability to offer empirical data on the population-wide impact of inherited Hb disorders. Despite growing awareness of haemoglobinopathies, many countries still lack structured surveillance systems. Where such data exist, they often remain underutilized due to fragmentation or insufficient integration into health planning3. In high-burden regions, such as sub-Saharan Africa, the value of Hb phenotype surveillance holds even greater significance. Yet comprehensive and sustained surveillance efforts are often undermined by infrastructural constraints, sociocultural barriers, and competing public health priorities. Consequently, affected individuals continue to face diagnostic delays, limited access to disease-modifying treatments, psychosocial challenges4, suboptimal quality of life and struggles with daily activities5. Surveillance data can bridge this gap, aiding in resource allocation and health system strengthening. This manuscript presents a ten-year retrospective analysis of Hb phenotype surveillance data, encompassing multiple regions and demographic strata using data from a large Nigerian university that attracts students from every zone in the country. The objectives are threefold: to quantify disease burden and its temporal trends, to assess carrier frequency variations across geographical and population groups where data permit, and to examine the broader public health and policy implications of these findings. In doing so, these findings expand available data by offering longitudinal insights into the epidemiology of haemoglobinopathies and proposing strategic recommendations for health system responses. Understanding the dynamics of Hb phenotype prevalence in such settings is vital to framing effective policy responses and designing interventions tailored to local contexts. Methods Study site and participants Between January 2015 and December 2024, a retrospective cohort dataset was analyzed using de-identified data from Bowen University Hospital’s [Iwo (population: 250,000–499,999), Osun State, southwestern Nigeria] health records unit. Ethical approval was sought and obtained from the Bowen University’s Directorate of Research and Strategic Partnerships. Dataset acquisition Data on haemoglobin phenotypes were collected from students during a ten-year period (2015–2024). Haemoglobin variant determination was performed by qualified Medical Laboratory Scientists using cellulose acetate haemoglobin electrophoresis at alkaline pH, which is routinely employed for haemoglobin variant screening in Nigeria6,7. Approximately 3 mL of venous blood was collected into EDTA anticoagulant tubes and processed according to standard laboratory procedures. Red blood cells were lysed to prepare haemolysates, which were then applied to cellulose acetate membranes pre-soaked in tris–EDTA–borate (TEB) buffer (pH 8.6). Electrophoresis was conducted at 250–300 V for approximately 15–20 min, allowing separation of haemoglobin fractions based on differences in their electrophoretic mobility. Following migration, the haemoglobin bands were visualized and compared with known control samples run concurrently. Haemoglobin variants were assigned according to the characteristic banding patterns observed (e.g., HbAA, HbAS, HbAC, HbSS, and HbSC). Internal quality control procedures were maintained throughout the study period by including control samples with established haemoglobin phenotypes in each analytical batch8. Data preprocessing Pre-processing became necessary due to missing and inconsistent values in different features. Student measurement values constituted most missing data, requiring data-cleaning techniques. We fixed errors, duplicates, outliers, and missing values9 through imputation (for missing values < 5%), duplicate removal, and transformation. Little’s MCAR test (p > 0.05) showed random missingness. Dataset cleaning yielded standardized ID, sex, genotype, and year fields. We grouped phenotypes into three clinical categories: Normal (HbAA), Carrier (HbAS and HbAC), and Disease (HbSS, HbSC, and HbCC). This classification aligns with clinical standards and improves data interpretability10,11. Software and statistical environment All analyses were performed in Python 3.11.4 using Jupyter Lab. Data handling and tabular summaries used pandas 2.0.3 and NumPy 1.24.3. Statistical testing (chi-square tests of independence, Pearson correlation, and ordinary-least-squares linear regression for annual trends) was implemented using SciPy 1.11.1 (scipy.stats). Clustering was implemented using scikit-learn 1.3.0. Prior to k-means clustering, the three input features (carrier rate, disease rate, and year) were standardized to zero mean and unit variance using scikit-learn’s StandardScaler. K-means was implemented with the k-means + + initialization method, 10 random re-initializations (n_init = 10), a maximum of 300 iterations (max_iter = 300), a convergence tolerance of 1 × 10⁻4 (tol = 1e−4), and a fixed random seed (random_state = 42) to ensure reproducibility. All figures were produced using Matplotlib 3.7.2 and Seaborn 0.12.2. All analysis code is publicly available at the repository referenced in the Availability of data and materials statement, with a version-controlled snapshot archived at Zenodo (https://doi.org/10.5281/zenodo.20451392). Data correlation and clustering For pattern discovery, Pearson correlation analysis and k-means clustering were applied. K-means clustering was used to explore temporal patterns in the yearly aggregated phenotype data rather than to cluster individual participants. Three features were used: carrier rate, disease rate, and year12. Details of the computational implementation (software versions, feature standardization, and the full k-means configuration, including initialization, number of restarts, maximum iterations, and convergence tolerance) are given in the preceding subsection. The optimal number of clusters (k = 3) was determined using the elbow method (within-cluster sum of squares) and silhouette analysis. Within-cluster sums of squares were 17.29 at k = 2, 9.23 at k = 3, and 6.51 at k = 4; silhouette scores were 0.34, 0.36, and 0.30 respectively, together supporting k = 3 as the most appropriate choice13,14. Colour coding was used in the clustering visualizations to distinguish temporal groups. Statistical significance was assessed at the 95% confidence level (p ≤ 0.05). Features correlation Pearson correlation values are reported in Table 1. To complement the correlation analysis, we applied ordinary-least-squares linear regression to assess annual trends for each phenotype and clinical category, and chi-square tests of independence to test for an association between sex and genotype distribution (χ2 = 1.49, df = 5, p = 0.91) and between sex and clinical category (χ2 = 0.39, df = 2, p = 0.82); none reached statistical significance (all p > 0.05), consistent with a stable pattern over time. The trend analyses likewise showed no statistically significant year-on-year change: a non-significant upward trend was observed for HbAA (slope = + 0.31 percentage points/year; R2 = 0.16; p = 0.25); a non-significant downward trend for the combined carrier category (slope = − 0.28 percentage points/year; R2 = 0.13; p = 0.31); and a non-significant downward trend for the combined disease (HbSS + HbSC + HbCC) category (slope = − 0.03 percentage points/year; R2 = 0.07; p = 0.45).Table 1Pearson correlation coefficients.VariablesCorrelation (r)p-valueGenotype × Gender0.0410.60Genotype × Year0.1030.34AA × Year0.1540.25AS × Year− 0.2380.51AC × Year− 0.5220.12 Table 1Pearson correlation coefficients.VariablesCorrelation (r)p-valueGenotype × Gender0.0410.60Genotype × Year0.1030.34AA × Year0.1540.25AS × Year− 0.2380.51AC × Year− 0.5220.12 Pearson correlation coefficients. Results Sociodemographic and medical records for the students are summarized in Tables 2, 3 and supplementary material. The analyses included 8,890 datasets gathered from students across various academic units. The study participants had a mean age of 16.82 years (SD = 1.67; range: 14–42). HbAA was the most common genotype (75.11%), followed by HbAS (20.40%), and there was a slight female predominance (57.6%, n = 5118).Table 2Gender distribution.SexCountPercentageFemale511857.57Male377242.43Table 3Overall genotype distribution.GenotypeCountPercentageAA667775.11AC2923.28AS181420.40CC20.02SC420.47SS630.71 Table 2Gender distribution.SexCountPercentageFemale511857.57Male377242.43 Gender distribution. Table 3Overall genotype distribution.GenotypeCountPercentageAA667775.11AC2923.28AS181420.40CC20.02SC420.47SS630.71 Overall genotype distribution. Year-to-year variability among phenotype percentages existed, though the relative distributions were consistent. In Fig. 1, HbAA reached its lowest point in 2016 (70.4%) and peaked during 2017 (77.8%). Carrier state (HbAS) showed opposite pattern – highest in 2016 (24.8%) and lowest in 2017 (16.7%). Disease types ranged from 0.5% in 2020 to 1.7% in 2018 as shown in Fig. S1.Fig. 1Distribution of genotypes by year (2015–2024). Distribution of genotypes by year (2015–2024). The phenotype distribution by sex is stable across all years. There were no significant gender differences among the major genotypes HbAA, HbAS, and HbAC, as shown in Fig. S2. Discussion This research screening for haemoglobinopathies among young adults, such as university students in Nigeria, a sickle cell hotspot, provides valuable insights into genotype prevalence patterns and changing genotype demographics in the country through longitudinal surveillance data15. Our study revealed that HbAA accounted for approximately three-quarters of the population, consistent with the previously documented national trends10,11,15–17. However, this finding is slightly higher than findings from North Central Nigeria, where only 53% had the normal HbAA genotype18, suggesting regional genetic diversity. This study’s prevalence of HbAS (sickle cell trait) at 20.4% aligns with the expected national range of 20–30%, though lower than the 23.4%, 24.6%, 32.7% and 40.57% reported in some studies7,18,19. The persistence of HbAS reinforces the evolutionary advantage of malaria resistance among carriers20,21, while underscoring the importance of routine screening, genetic counselling, and awareness to prevent SCD in future generations. The prevalence of HbAC was 3.29%, consistent with recent national reports15 but higher than the findings of Ohiengbomwan et al.19 highlighting the genotype’s lower but clinically relevant burden. Although individuals with HbAC are largely asymptomatic, their mild haemoglobinopathy profile and partial malaria protection warrant consideration in screening programs. SCD genotypes (HbSS and HbSC) appeared in just 1.18% of the study population, lower than national estimates of 1.5–3%,22 and also differ from the findings of Ameen et al.18 and Azi et al.23 This relatively low prevalence may reflect the adverse impact of SCD on childhood survival, educational attainment, and university admission, i.e. those with SCD may not be surviving or thriving well enough to attend the university. Despite its low occurrence in this group, SCD remains a critical health concern, especially in resource-limited settings where it contributes significantly to morbidity and mortality. Trends over time (2015–2024) revealed a remarkably consistent haemoglobin variant distribution, though minor changes were observed. The disease rate remained relatively stable, fluctuating around 1.2% while the normal (HbAA) category consistently comprises about 75% of the sample population. The lack of statistical significance in these trends demonstrates stable phenotype distribution throughout our study. This stability of phenotype distribution across the course of the investigation supports a persistent public health need for ongoing phenotype screening for haemoglobinopathies and awareness campaigns. With respect to public health implications, the relatively stable carrier prevalence (~ 20–25%) across the decade highlights the ongoing danger of sickle cell disease births in the absence of adequate genetic counseling and premarital screening programmes. These findings underline the importance of better national screening strategies and sustained public health education. K-means clustering analysis grouped the 10-year surveillance data into three clusters (Table 4 and Fig. S3), interpretable as follows:Period/Cluster 1 (2020, 2022, 2023): A “stabilization period” characterized by moderate carrier rates (24.1%) and the lowest disease burden (0.9%), suggesting improved population health outcomes.Period/Cluster 2 (2017, 2019, 2021, 2024): A “transition period” showing the lowest carrier rates (21.9%) but paradoxically higher disease prevalence (1.4%), indicating possible changes in screening practices or population dynamics.Period/Cluster 3 (2015–2016, 2018): A “high-burden period” exhibiting both the highest carrier rates (25.7%) and disease prevalence (1.5%), representing the baseline epidemiological state.Table 4Temporal clustering of epidemiological periods.Period/clusterYearsCarrier rate (%)Disease rate (%)Characteristics12020–202324.1 (23.1–25.3)0.9 (0.4–1.2)Stabilization period22017, 2019, 2021, 202421.9 (20.8–23.0)1.4 (1.0–1.8)Transition period32015–2016, 201825.7 (23.3–29.1)1.5 (1.1–1.8)High-burden period Period/Cluster 1 (2020, 2022, 2023): A “stabilization period” characterized by moderate carrier rates (24.1%) and the lowest disease burden (0.9%), suggesting improved population health outcomes. Period/Cluster 1 (2020, 2022, 2023): A “stabilization period” characterized by moderate carrier rates (24.1%) and the lowest disease burden (0.9%), suggesting improved population health outcomes. Period/Cluster 2 (2017, 2019, 2021, 2024): A “transition period” showing the lowest carrier rates (21.9%) but paradoxically higher disease prevalence (1.4%), indicating possible changes in screening practices or population dynamics. Period/Cluster 2 (2017, 2019, 2021, 2024): A “transition period” showing the lowest carrier rates (21.9%) but paradoxically higher disease prevalence (1.4%), indicating possible changes in screening practices or population dynamics. Period/Cluster 3 (2015–2016, 2018): A “high-burden period” exhibiting both the highest carrier rates (25.7%) and disease prevalence (1.5%), representing the baseline epidemiological state. Period/Cluster 3 (2015–2016, 2018): A “high-burden period” exhibiting both the highest carrier rates (25.7%) and disease prevalence (1.5%), representing the baseline epidemiological state. Table 4Temporal clustering of epidemiological periods.Period/clusterYearsCarrier rate (%)Disease rate (%)Characteristics12020–202324.1 (23.1–25.3)0.9 (0.4–1.2)Stabilization period22017, 2019, 2021, 202421.9 (20.8–23.0)1.4 (1.0–1.8)Transition period32015–2016, 201825.7 (23.3–29.1)1.5 (1.1–1.8)High-burden period Temporal clustering of epidemiological periods. These temporal clusters demonstrate that, despite overall genetic stability, subtle but meaningful variations occurred across the decade, with disease rates ranging from 0.4 to 1.8% and carrier rates from 20.8 to 29.1%. Three-dimensional clustering visualization (Fig. S3) confirmed clear separation between these temporal periods, revealing underlying heterogeneity in population genetics across the study period. We note that the cluster labels above (“stabilization,” “transition,” and “high-burden”) are post-hoc interpretive descriptions of the algorithm’s output and are not intended as formally defined epidemiological phases. The identification of three distinct epidemiological periods through our longitudinal analytic approach provides novel insights into the temporal dynamics of SCD in our population. While we did observe some stability of genetic distributions over time, important nuances came out of our clustering approach. For example, the “stabilization period” (2020, 2022, 2023) showed lower disease rates, whereas the “transition period” pattern, where lower carrier rates accompany higher disease rates, suggests complex population dynamics that warrant further investigation. These temporal patterns challenge the assumption of universal static genetic equilibrium and highlight the value of longitudinal surveillance combined with advanced analytical approaches. This research enhances phenotype surveillance frameworks in three significant ways: First, it indicates the feasibility of establishing a decade-long institutional surveillance system in a resource-constrained setting. Second, it proposes a methodological framework (using clustering analysis to identify discrete epidemiological patterns and periods) to supplement traditional trend analysis in surveillance programmes. Third, it provides empirical data that can inform national SCD policy, including setting targets for screening programmes and allocating resources for afflicted persons. Phenotype surveillance can be used to actively promote policy and equity by allowing for targeted resource allocation, informing premarital screening programs, identifying marginalized communities, and assisting in the implementation of national SCD control plans. When combined with community participation, such approaches can enhance fair access to genetic counseling while also reducing disease stigma. These datasets were obtained from long-term data collection of university student cohorts. As several nations have demonstrated, phenotype surveillance could support national SCD programmes. Countries such as Bahrain and Saudi Arabia have implemented premarital screening programs that significantly reduced at-risk marriages through informed reproductive choices24. These examples suggest that genetic surveillance, when integrated with legal frameworks, community sensitization, and accessible genetic counselling services, may contribute meaningfully to public health outcomes. However, surveillance alone is insufficient without community engagement and cultural sensitivity. Misconceptions about haemoglobinopathies, ranging from spiritual interpretations to outright stigma, persist in many affected communities. No matter how comprehensive, surveillance data can be underutilized or mistrusted if the populations involved do not understand or accept the rationale behind testing25. While we feel our data support educational outreach strategies, they must be done in a culturally-literate manner to be effective. Our data offer longitudinal phenotype prevalences for young adults at a large Nigerian academic institution; however, because the cohort consists of admitted university students, a relatively healthy and educated young adult population, these findings may not reflect population-level prevalence and should be interpreted within that restricted context. However, the retrospective data analyses that we pursued does have limitations. Given the retrospective nature, we are unable to make causal observations about phenotype variances. We also are using a university cohort, and there is strength in the total number of observed data and the longitudinal nature, but whether there are varying trends in young adults who do not attend the University is uncertain. Specifically, the use of a university-based cohort in this study may introduce selection bias, as individuals with severe sickle cell disease may be inadequately represented due to diminished survival rates and lower educational achievement. Consequently, prevalence rate of 1.18% observed in this study may underestimate the true burden of SCD in the general Nigerian population. The absence of geographic identifiers hindered our capacity to adequately examine regional changes in phenotype distribution, thereby confining the interpretation of population-level variability. Despite these limitations, by drawing from a decade’s worth of surveillance data, this work offers a unique vantage point from which to evaluate the successes, shortcomings, and future directions of haemoglobinopathy control efforts. Conclusion Universities provide effective settings for haemoglobinopathy screening because they can represent cultural mixing that may not occur in regional communities and offer insights into the next child-bearing generation when it comes to genetic disorders. Our analyses found some level of stability in Hb phenotypes but also discovered nuanced trend groups that were more dynamic than expected. By identifying individuals with SCD-associated genotypes who need specialized healthcare and demonstrating long-term data trends, our observations and analyses can be used to aid institutional healthcare planning and resource allocation, but it can also go beyond just the medical space. Since university populations represent not only the next generation to have families but also the next generation of leaders and policy developers, analyzing longitudinal phenotype data may help expand policy proposals and increase opportunities for healthcare improvements to reduce disease burden at the regional and national levels. These findings should be interpreted within the context of a single-institution cohort of admitted university students and may not reflect population-level prevalence. Within that limitation, our study provides a basis for institutional genotype surveillance that, if replicated across several universities and healthcare institutions, may inform the development of a broader national haemoglobinopathy surveillance approach. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1