Multilevel analysis of risk factors for clinical mastitis in dairy cows in Plateau State, Nigeria: a hierarchical mixed-effects logistic regression modelling approach Authors: Joshua Onyango, Aminu Shittu, Mohammad-Bashir Bolajoko, Folorunso O Fasina, Andrew Peters Published: 2026 ABSTRACT AbstractMastitis remains a significant health challenge affecting dairy cows, with implications for animal welfare, milk production, and farm profitability. In Nigeria, there are limited large-scale studies that have systematically investigated the multifactorial risk factors for mastitis across diverse production systems and geographical locations. This study aimed to assess the prevalence of mastitis and identify its risk factors using a hierarchical mixed effects logistic regression, block-wise analytical approach. A cross-sectional survey was conducted across 298 dairy farms in North-Central Nigeria. Data were collected on farm demographics, animal characteristics, housing and management practices, water and feeding routines, milking hygiene, and mastitis management. Mastitis was defined at the herd level as the occurrence of at least one farmer-reported case within the current lactation cycle. Because mastitis diagnosis relied on farmer-reported cases rather than laboratory confirmation (e.g., California Mastitis Test, CMT, or somatic cell count, SCC), the possibility of misclassification bias cannot be excluded and should be considered when interpreting the findings. Univariable and multivariable mixed-effects logistic regression models were fitted for each conceptual block of variables, accounting for clustering at the local government area (LGA) level using random intercepts. Model performance was evaluated using likelihood ratio tests, intraclass correlation coefficients (ICCs), and diagnostic plots of residuals. The prevalence of reported mastitis in the current year was 59.4% (95% CI: 53.7–64.9). Final multivariable models revealed significant associations between mastitis and herd size, presence of working bulls, milking hygiene (e.g., teat dipping and use of separate cleaning cloths), and mastitis treatment practices. Notably, large cattle herds had significantly higher odds of mastitis (adjusted OR = 6.56, 95% CI: 1.99–21.62), while post-milking teat dipping (OR = 0.013, 95% CI: 0.000–0.797) was strongly associated with lower odds. The ICC values across models ranged from 0.58 to 0.83, indicating substantial variation at the LGA level. Mastitis is highly prevalent in Nigerian dairy farms, with multivariable risks: herd demographics, management practices, and hygiene behaviours. Interventions promoting evidence-based milking hygiene and targeted herd-level management could substantially reduce mastitis burden. The hierarchical modeling approach provides a comprehensive framework for identifying context-specific risk factors and guiding regionally appropriate control strategies.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12917-026-05643-0.Keywords: Mastitis, Dairy cows; Mixed-effects logistic regression; Risk factors; Milking hygiene; Nigeria; Herd management; Disease prevalence Introduction Mastitis is one of the most economically significant diseases affecting dairy cattle worldwide. It is characterized by inflammation of the mammary gland, commonly caused by bacterial infection, and manifests in clinical or sub-clinical forms [1, 2]. The disease not only affects milk yield and quality, but also poses public health risks due to possible antimicrobial residues in milk and the transmission of zoonotic pathogens such as Staphylococcus aureus, Escherichia coli, and Streptococcus agalactiae, which have documented public health implications through contaminated milk [3–5]. Sub-clinical mastitis, in particular, often goes unnoticed but has long term implications for cow health, reproductive performance, and farm profitability [6, 7]. In many smallholder dairy systems, mastitis detection often relies on farmer recognition of clinical signs due to limited access to laboratory diagnostic tools [8–13]. In sub-Saharan Africa (SSA), including Nigeria, the burden of mastitis is exacerbated by inadequate disease surveillance, limited veterinary infrastructure, and poor adherence to hygienic milking practices [14]. Various studies conducted in Nigeria have documented high prevalence rates of mastitis, especially in the sub-clinical form. For instance, a study by Shittu et al. [15] in the Savannah Region of Nigeria found a significant burden of sub-clinical mastitis among lactating cows, with poor milking hygiene and inadequate disease management identified as major risk factors. Other regional studies corroborate these findings, highlighting the widespread nature of the disease and its link to poor management practices [16–18]. Despite the growing body of literature on mastitis in Nigeria, most studies have focused on either descriptive epidemiology or small-scale investigations within specific farms or abattoirs. Few have systematically examined herd and management level risk factors using robust multilevel statistical models. Importantly, while several studies in Nigerian dairy systems have documented mastitis prevalence and management practices, many have relied primarily on descriptive analyses or single-level statistical approaches. These approaches often do not account for the hierarchical structure of dairy production systems, where farm management, environmental conditions, and regional factors may cluster and jointly influence disease occurrence. The present study addresses this methodological gap by applying a hierarchical mixed-effects modelling framework that explicitly accounts for clustering and multilevel risk structures in mastitis epidemiology. In addition, the interplay between socio-demographic variables, housing systems, feeding practices, and herd-level mastitis status remains poorly understood. There is also limited evidence on the effectiveness of common mastitis control practices in typical smallholder dairy settings, where resource constraints and traditional knowledge often shape farm decisions. Given the economic and public health implications of mastitis, there is a pressing need for comprehensive, farm-level studies that integrate multiple domains of risk using advanced analytical techniques. Plateau State, situated in the North-Central highlands of Nigeria, represents a unique setting where smallholder dairy production intersects with evolving management systems. Understanding the determinants of mastitis in this context is essential for designing targeted interventions and improving dairy productivity in the region. The objectives of this study were to estimate the herd-level prevalence of reported clinical mastitis in dairy farms and to identify farm-level, environmental, and herd structural factors associated with mastitis occurrence at the herd level. In addition, the study aimed to determine significant predictors of mastitis using a multilevel hierarchical multivariable mixed-effects logistic regression modeling framework, accounting for clustering effects within the study area. Materials and methods Study area This study was conducted in Plateau State, located in the North-Central region of Nigeria. Plateau State is characterized by a relatively temperate highland climate compared with many other parts of the country, owing to its elevated topography. The state covers approximately 30,913 km² and comprises 17 LGAs. Climatic conditions in the region typically include average temperatures ranging between about 18 °C and 28 °C, with a distinct rainy season occurring approximately from April to October and a dry season from November to March. These seasonal patterns influence forage availability, water resources, and livestock management practices, which may in turn affect disease dynamics in dairy herds. The region consists largely of rural and peri-urban agricultural communities where dairy farming, predominantly practiced by smallholder farmers, represents an important livelihood activity. Livestock production systems in the state range from extensive pastoral systems to semi-intensive and intensive management systems. A detailed list of localities and their corresponding LGAs was obtained from a publicly accessible database to support data verification and geographic coding [19]. Study design and population A cross-sectional study design was employed to collect data from dairy farms and households across selected LGAs in Plateau State. The study population comprised lactating dairy cows and their respective farmers. Farms constituted the primary analytical unit in this study, and mastitis status was defined at the herd level based on the occurrence of at least one farmer-reported case of clinical mastitis within the herd during the study period. Inclusion criteria required farms to have at least one lactating cow at the time of data collection. Households not engaged in dairy production or lacking essential data were excluded from the final analysis. Sampling strategy A multi-stage sampling approach was adopted for farm selection. In thefirst stage, LGAs with active dairy farming activities were purposively selected to ensure representation of the major dairy production systems present in Plateau State, including extensive pastoral, semi-intensive, and emerging peri-urban dairy systems. This approach was used to capture the diversity of management practices and herd structures within the region rather than to achieve strict probability sampling across all LGAs. The selection of LGAs was informed by expert consultation with veterinary extension officers and preliminary field reconnaissance. This approach enabled the inclusion of areas with substantial dairy production; however, purposive selection may introduce some degree of selection bias and could limit the full representativeness of the sampled LGAs within Plateau State. This limitation was considered when interpreting the generalizability of the findings. In the second stage, dairy farms and households within the selected LGAs were randomly sampled with the assistance of local livestock extension officers to ensure adequate coverage of dairy-producing communities. The sample size was determined using the standard epidemiological formula for prevalence studies, assuming a 50% expected prevalence of mastitis (to maximize sample size), a 95% confidence level, and a 5% margin of error, with adjustments for clustering and non-response. Where: n= required sample size. n= required sample size. Z= Z-score for 95% confidence (1.96). Z= Z-score for 95% confidence (1.96). p= expected prevalence of mastitis (0.5 used for maximum variability). p= expected prevalence of mastitis (0.5 used for maximum variability). d= desired margin of error (0.05). d= desired margin of error (0.05). The calculated sample size was subsequently adjusted to account for the study design effect arising from clustering and potential non-response. Data collection Clinical mastitis in cows was identified based on farmer-reported cases occurring during the current lactation cycle. Farmers reported mastitis based on observable clinical signs including abnormal milk (e.g., watery, bloody, or containing flakes, clots, or pus) as well as visible udder abnormalities such as swelling, heat, redness, and pain in the affected quarter of the cow’s udder. A farm was classified as mastitis-positive if the farmer confirmed that at least one lactating cow had exhibited any of these clinical signs during the study period. This herd-level classification based on the presence of at least one clinical case is consistent with epidemiological approaches used in farm-level studies, where the objective is to identify herd management and environmental risk factors rather than estimate individual animal incidence. Although diagnostic confirmation methods such as the CMT or SCC were not used, these tests are not routinely practiced among smallholder dairy farmers in the study area. In many smallholder dairy systems, farmer recognition of visible clinical signs is commonly used for mastitis detection under field conditions. Previous studies have shown that farmer-reported mastitis can serve as a practical proxy indicator of clinical mastitis in epidemiological investigations where laboratory diagnostics are not readily available [8–13, 20–24]. Nevertheless, the reliance on farmer-reported cases may introduce some degree of misclassification due to variability in farmers’ ability to recognize mastitis symptoms, and this limitation was considered in the interpretation of the study findings. Related data were collected using a structured questionnaire administered by trained enumerators (Supplementary material S1). The questionnaire was developed based on existing literature and expert input and pretested in a pilot study to ensure clarity and relevance Supplementary material S1. It captured demographic characteristics of the farmers, herd structure, housing and management practices, water and feed sources, milking hygiene, and mastitis control practices. The primary herd-level mastitis status variable was the binary response to whether at least one case of clinical mastitis occurred in the herd during the current lactation cycle (Mastitisthisyear: 1 = Yes, 0 = No). Ethical clearance was obtained from the National Veterinary Research Institute, Nigeria, ethics board, and informed consent was sought and received from all participants. Data management and statistical analysis All collected data were entered into Microsoft Excel and exported to STATA/IC 15.0 [25], for cleaning and analysis. Initial data validation included correcting misspelled entries and harmonizing inconsistent location names by cross-checking with the official list of districts and villages in Plateau State. Localities with sparse observations (e.g., fewer than 10 farmers) were aggregated under their corresponding LGAs to improve analytical robustness. For instance, entries such as “Bwall/Quaanpan,” “Kwalla/Quaanpan,” and “Kwang/Quaanpan” were all reclassified under “Qua’an Pan LGA.” Farmers who indicated “Inherited” for dairy experience were matched with their reported biological ages under the assumption that they had lifelong exposure to dairy farming. Responses of “family” or “0” under the number of employees variable were recoded as “0,” indicating no hired labour. Under “Housing design,” rare responses like “Extensive” and “Open” were combined into an “Other” category due to conceptual overlap and low frequency. The variable representing herd size (total number of animals) was collapsed into a binary classification: “small herds” (1–15 animals or missing values) and “large herds” (> 15 animals), in line with the observed distribution and practical herd management differences. Continuous or highly granular categorical variables were grouped into broader categories where necessary to ensure adequate observations within each category and to facilitate meaningful epidemiological interpretation. Such categorization also improves model stability and interpretability in regression analyses. The “Water source” variable was harmonized into five categories: pipeline, well/borehole, lake/dam, river/stream, and other. Similarly, “Type of feed” was recoded into concentrate, blocks, grazing, and other, with ambiguous or textual responses grouped accordingly. Other continuous variables like age and years of experience were categorized using quartiles and meaningful age bands (e.g., 18–29, 30–39, 40–49, 50–59, 60 + years) to support stratified analysis. Missing values were handled through a combination of logical verification and data cleaning procedures. During data entry and validation, inconsistent or incomplete entries were cross-checked against the original field records where possible. Where values could not be logically verified or reliably reconstructed, the affected observations were excluded from analyses for the respective variables to maintain data integrity. A hierarchical block-wise modeling approach was used for multivariable analysis. Explanatory variables were grouped into six conceptual blocks: (1) Farmer and Farm-Level Demographics, (2) Animal Demographics and Structure, (3) Housing and Management Practices, (4) Water and Feed, (5) Milking Hygiene Practices, and (6) Mastitis Management. Within each block, univariable mixed-effects logistic regression models were first fitted, accounting for clustering at the LGA level using random intercepts. Because the primary analytical unit of the study was the farm (herd), mastitis status was defined at herd level based on the presence of at least one reported case within the lactation cycle. To account for potential geographical clustering of farms within the same administrative area, mixed-effects logistic regression models were fitted with random intercepts at the LGA level. Variables with a p-value < 0.25 were selected for further multivariable analysis. A relatively liberal screening threshold was intentionally used at this stage to avoid excluding variables that may not appear significant in univariable analysis but could become important predictors or confounders in the multivariable model. This approach follows established epidemiological modelling guidance recommending the use of a higher screening threshold during preliminary variable selection [26–28]. The organization of these blocks and the specific inclusion criteria for each variable are detailed in Appendix Table 1. This table outlines the six thematic blocks used in the hierarchical multivariable mixed-effects logistic regression modeling process. Each block contains conceptually related variables selected to evaluate their individual and combined associations with the occurrence of mastitis in the current year. The choice of hierarchical modelling framework and the developed directed acyclic graph was based on biological and epidemiological reasoning, and reflective of the natural causal chain of mastitis occurrence earlier proposed [29, 30] (Fig. 1). Such explanatory modeling, incorporating biological parameters and epidemiological understanding, ensures that the relationships among risk factors mirror the pathophysiological and management dynamics of mastitis in real-world systems (Fig. 1). Directed Acyclic Graph (DAG) representing Causal Pathways for Clinical Mastitis. These causal pathways and clarify model hierarchy strengthen the theoretical coherence and interpretability of the statistical framework The general form of the first model equation (mixed-effects logistic model) was: Where: is the binary herd-level mastitis status for farm i in LGA j (1 = mastitis reported, 0 = none reported). is the binary herd-level mastitis status for farm i in LGA j (1 = mastitis reported, 0 = none reported). is the model intercept. is the model intercept. is the regression coefficient for predictor X. is the regression coefficient for predictor X. is the random intercept for LGA j, where u . is the random intercept for LGA j, where u . Variables with p < 0.25 were retained for multivariable modeling. Multivariable mixed effects logistic regression was performed using backward stepwise deletion [31]. Although stepwise procedures were used to assist model simplification, variable selection was not based solely on statistical criteria. Variables retained in the final models were also evaluated based on biological plausibility, epidemiological relevance, and potential confounding effects. Model diagnostics and goodness-of-fit measures were additionally assessed to ensure that the final models were both statistically robust and epidemiologically interpretable. Variables were sequentially removed if they did not significantly improve the model fit, as evaluated using the likelihood-ratio test (p < 0.05 as the threshold). Nested models were compared, and the best fitting models were selected based on likelihood ratio test results, Akaike’s Information Criterion (AIC), and Bayesian Information Criterion (BIC). AIC and BIC were used to evaluate the relative goodness-of-fit of competing models while accounting for model complexity, with lower values indicating a better balance between explanatory power and model parsimony [28]. AIC and BIC were used to compare competing mixed-effects models fitted to the same dataset and outcome structure. These criteria are appropriate for evaluating relative model fit while penalizing model complexity, particularly in hierarchical modelling frameworks where overfitting may arise from excessive parameterization. The final models for each block were combined to develop a comprehensive model. It should be noted that the choice of variable selection process followed a widely accepted univariable screening criterion (p < 0.25) [26–28] (Appendix Fig. 1). Confounders were retained based on a ≥ 10% change in coefficients rather than significance alone. Multicollinearity among explanatory variables was assessed using Variance Inflation Factors (VIF), calculated from the variance–covariance matrix of the regression coefficients. VIF values below 2 were considered indicative of absence of problematic multicollinearity among predictors. Intraclass correlation and the Hosmer–Lemeshow goodness-of-fit test were also assessed to evaluate model adequacy and mitigate overfitting. The general form of the final multivariable model was: Where: represents the k-th predictor variable for farm i in cluster j. represents the k-th predictor variable for farm i in cluster j. represents the corresponding regression coefficients. represents the corresponding regression coefficients. represents the LGA-level random intercept. represents the LGA-level random intercept. Model comparison was based on: Likelihood-ratio test (LRT): Likelihood-ratio test (LRT): AIC: AIC: BIC: BIC: Where L= likelihood, k= number of parameters, n= sample size. The ICC was estimated to quantify the proportion of total variance attributable to differences between clusters (LGAs) [32]. Post-estimation diagnostics included calculating and plotting marginal effects using the margins and margins-plot commands. Pearson residuals were predicted to examine model adequacy [33]. A histogram of the residuals overlaid with a normal density curve was used to assess the distribution of residuals, while a scatter plot of residuals against predicted probabilities (fitted values) was used to evaluate heteroscedasticity and model misspecification [34]. Where is the variance of the standard logistic distribution. Marginal effects: Marginal effects: Residual analysis using Pearson residuals: Residual analysis using Pearson residuals: These were plotted in histograms and against predicted probabilities to assess model fit. All statistical analyses were conducted using STATA/IC 15.0, and findings were reported in accordance with established reporting guidelines for observational epidemiological studies [35]. Results Prevalence of mastitis among surveyed farms Out of 298 surveyed dairy farms, 177 reported at least one case of clinical mastitis in the current lactation cycle, yielding a herd-level prevalence of 59.4% (95% CI: 53.7–64.9). Herd-level mastitis was defined as the presence of at least one reported case within the lactation cycle (Supplementary material S2). In the univariable mixed-effects logistic regression analysis, farmer demographic characteristics – including age, education, farming experience, and number of employees – showed a statistically significant association (p < 0.05) with mastitis risk (Table 1). However, variables related to herd composition showed no statistically significant associations with mastitis. Larger numbers of milking cows (20–49; OR = 1.96, p = 0.158) and heifers > 2 years never calved (11–20; OR = 2.66, p = 0.094) were associated with higher estimated odds, while having few female calves (1–3; OR = 0.41, p = 0.059) was associated with lower estimated odds of mastitis, although the association was not statistically significant (Table 2). Similarly, intensive production systems were associated with lower odds of mastitis compared to extensive systems (OR = 0.155, p = 0.028). Large herd size (based on total cattle) was associated with higher estimated odds of mastitis, although the association did not reach statistical significance (OR = 2.209, p = 0.058) (Table 3). No water or feed variables reached statistical significance. Feeding cows after milking (OR = 0.435, p = 0.083) and grazing-based feeding (OR = 0.411, p = 0.141) were associated with lower estimated odds of mastitis, although these associations were not statistically significant (Table 4). Table 1Univariable mixed-effects logistic regression modeling for Block 1: farmer & farm-level demographicsVariableLevelOR (crude)SEzP> z95% CIAge group (years)30–391.00Ref.Ref.Ref.Ref.18–290.8420.45-0.320.7480.295 – 2.40040–490.8960.411-0.240.810.364 – 2.20350–592.0331.2211.180.2380.626 – 6.59960+1.3340.7940.480.6290.415 – 4.282EducationIlliterate1.00Ref.Ref.Ref.Ref.College/Diploma0.9911.014-0.010.9930.133 – 7.368High school2.4822.1021.070.2830.472 – 13.047Primary school0.5040.274-1.260.2080.173 – 1.465University1(empty)*Years in dairy farming16–301.00Ref.Ref.Ref.Ref.1–54.6065.8371.210.2280.384 – 55.1966–152.0571.1221.320.1860.706 – 5.99231–451.870.841.390.1630.776 – 4.50946+1.9291.2141.040.2970.562 – 6.625Number of employees1–21.00Ref.Ref.Ref.Ref.None0.9130.765-0.110.9140.177 – 4.7183–41.6810.8421.040.30.630 – 4.4855–102.0761.3821.10.2720.563 – 7.65211+4.7995.2221.440.150.569 – 40.493*Empty cells represent exposure categories with zero observations or combinations with no mastitis cases in the dataset Univariable mixed-effects logistic regression modeling for Block 1: farmer & farm-level demographics *Empty cells represent exposure categories with zero observations or combinations with no mastitis cases in the dataset Table 2Univariable mixed-effects logistic regression modeling for Block 2: animal demographics and structureVariableLevelOR (crude)SEzP> z95% CIMilking cowsSmall (2–9)1.00Ref.Ref.Ref.Ref.Medium (10–19)1.4130.6190.790.430.599 – 3.336Large (20–49)1.9590.9341.410.1580.770 – 4.988Very large (50+)1.0071.4530.010.9960.060 – 17.031Dry cows awaiting calvingSmall (1–5)1.00Ref.Ref.Ref.Ref.None (0)2.622.2891.10.270.473 – 14.518Medium (6–10)1.3950.5730.810.4170.624 – 3.119Large (11–20)1.2990.6630.510.6080.478 – 3.535Very large (>20)0.9150.832-0.10.9220.154 – 5.442Milking cows pregnantFew (1–4)1.00Ref.Ref.Ref.Ref.None (0)0.4470.25-1.440.150.149 – 1.339Some (5–9)1.7280.7761.220.2230.717 – 4.168Many (10–19)0.9580.563-0.070.9420.303 – 3.033Most/All (20+)0.5360.444-0.750.4520.105 – 2.721Milking cows not pregnantFew (1–4)1.00Ref.Ref.Ref.Ref.None (0)0.7060.303-0.810.4170.304 – 1.638Some (5–9)0.8530.452-0.30.7640.302 – 2.408Many (10–19)0.6080.437-0.690.4890.148 – 2.489Most/All (20+)0.6850.619-0.420.6750.117 – 4.022Heifers >2 yrs never calvedLow (1–5)1.00Ref.Ref.Ref.Ref.None (0)0.9160.707-0.110.910.202 – 4.161Moderate (6–10)1.1030.5520.20.8450.414 – 2.940High (11–20)2.6561.5481.680.0940.848 – 8.326Very high (>20)2.0041.8640.750.4550.324 – 12.409Heifers served awaiting confirmationLow (1–3)1.00Ref.Ref.Ref.Ref.None (0)1.4830.8340.70.4830.493 – 4.462Moderate (4–6)2.7781.5631.820.0690.922 – 8.371High (7–10)1.3440.9620.410.680.330 – 5.467Very high (>10)0.4940.359-0.970.3320.119 – 2.051Female calves <1 yearModerate (4–6)1.00Ref.Ref.Ref.Ref.None (0)0.2030.381-0.850.3960.005 – 8.093Few (1–3)0.4060.194-1.890.0590.159 – 1.036Many (7–10)0.7770.353-0.560.5780.319 – 1.891Very many (>10)0.6560.374-0.740.4590.215 – 2.005 Univariable mixed-effects logistic regression modeling for Block 2: animal demographics and structure Table 3Univariable mixed-effects logistic regression modeling of mastitis in relation to housing and management practicesVariableLevelOR (crude)SEzP> z95% CIHousing designOutdoor1.00Ref.Ref.Ref.Ref.Indoor2.9982.3591.40.1630.641 – 14.019Other2.0661.4241.050.2920.535 – 7.974Semi-indoor1(empty)*Cow accommodationGroup1.00Ref.Ref.Ref.Ref.Individual0.5960.433-0.710.4760.144 – 2.473Animals move out for grazingYes1.00Ref.Ref.Ref.Ref.No5.2236.1321.410.1590.523 – 52.157Production systemExtensive1.00Ref.Ref.Ref.Ref.Intensive0.1550.132-2.20.0280.030 – 0.817Semi-intensive0.4850.311-1.130.2580.138 – 1.702Smallholder0.220.221-1.510.1310.031 – 1.573Herd size (total animals)Large herds1.00Ref.Ref.Ref.Ref.Small herds0.2310.23-1.470.1420.033 – 1.631Herd size (total cattle)Medium herds1.00Ref.Ref.Ref.Ref.Small herds0.9410.473-0.120.9040.351 – 2.522Large herds2.2090.9221.90.0580.975 – 5.007Working bullsFew (1–2)1.00Ref.Ref.Ref.Ref.None1.2581.4240.20.8390.839 – 11.562Moderate (3–5)2.1681.0731.560.1180.821 – 5.720Many (6–10)1.1430.7640.20.8410.309 – 4.234Very Many (>10)0.3750.297-1.240.2150.079 – 1.768*Empty cells represent exposure categories with zero observations or combinations with no mastitis cases in the dataset Univariable mixed-effects logistic regression modeling of mastitis in relation to housing and management practices *Empty cells represent exposure categories with zero observations or combinations with no mastitis cases in the dataset Table 4Univariable mixed-effects logistic regression modeling for Block 4: water and feed practicesVariableLevelOR (crude)SEzP> z95% CIWater sourceRiver/Stream1.00Ref.Ref.Ref.Ref.Lake/Dam0.9290.606-0.110.9110.259 – 3.334Other1.1510.8220.20.8440.284 – 4.669Pipeline0.8951.385-0.070.9430.043 – 18.592Well/Borehole3.7233.8311.280.2010.496 – 27.975Water ad libitum?Yes1.00Ref.Ref.Ref.Ref.No0.9840.402-0.040.9690.442 – 2.192Feed supplements?No1.00Ref.Ref.Ref.Ref.Yes0.8570.444-0.30.7660.311 – 2.364Type of feedOthers1.00Ref.Ref.Ref.Ref.Blocks2.1633.4550.480.6290.094 – 49.528Concentrate0.6640.477-0.570.5690.163 – 2.713Grazing0.4110.248-1.470.1410.125 – 1.344Individual or group feeding?Group1.00Ref.Ref.Ref.Ref.Individual0.4680.566-0.630.530.044 – 5.012Supplement to lactating cows?Lactating & dry1.00Ref.Ref.Ref.Ref.None0.1320.179-1.490.1360.009 – 1.894Only dry3.3123.3961.170.2430.444 – 24.711Only lactating0.4690.375-0.950.3440.098 – 2.250Other2.2481.41.30.1940.663 – 7.621Feed cows after milking?No1.00Ref.Ref.Ref.Ref.Yes0.4350.209-1.740.0830.170 – 1.114 Univariable mixed-effects logistic regression modeling for Block 4: water and feed practices Several hygiene practices were significantly associated with lower odds of mastitis: teat dipping/spraying before milking (OR = 0.137, p = 0.012) and after milking (OR = 0.107, p = 0.011), as well as cleaning only visibly dirty teats (OR = 0.321, p = 0.011) (Table 5). Table 5Univariable mixed-effects logistic regression modeling for Block 5: milking hygiene practicesVariableLevelOR (crude)SEzP> z95% CIMilkerFamily member1.00Ref.Ref.Ref.Ref.Farmer0.750.32-0.680.4990.325 – 1.729Handwash/wear glove during milking?No1.00Ref.Ref.Ref.Ref.Yes0.6390.299-0.960.3390.255 – 1.601Clean teats as preparation for milking?No1.00Ref.Ref.Ref.Ref.All cows0.2240.191-1.750.080.042 – 1.195Only dirty teats0.3210.143-2.550.0110.134 – 0.768Dry teat before milking?No1.00Ref.Ref.Ref.Ref.Yes1.2190.6740.360.7210.412 – 3.605Use separate clothes for cleaning?No1.00Ref.Ref.Ref.Ref.Yes0.8580.414-0.320.750.333 – 2.210Pre-strip before milking?No1.00Ref.Ref.Ref.Ref.Yes0.8770.428-0.270.7880.337 – 2.282Teat dip/spraying before milking?No1.00Ref.Ref.Ref.Ref.Yes0.1370.109-2.50.0120.029 – 0.650Teat dip/spraying after milking?No1.00Ref.Ref.Ref.Ref.Yes0.1070.094-2.540.0110.019 – 0.601Milking methodHand1.00Ref.Ref.Ref.Ref.Machine0.1850.295-1.060.290.008 – 4.213Frequency of washing milking equip.Daily1.00Ref.Ref.Ref.Ref.Monthly4.4763.9121.710.0860.807 – 24.824Weekly0.2340.27-1.260.2090.024 – 2.258Servicing milking equipmentWhen broken1.00Ref.Ref.Ref.Ref.Annually2.771.7051.650.0980.829 – 9.256Monthly1.6080.9780.780.4350.488 – 5.296 Univariable mixed-effects logistic regression modeling for Block 5: milking hygiene practices The variables, ‘not treating mastitis’ (OR = 0.226, p = 0.003) and ‘reporting no recurrence of treated mastitis’ (OR = 0.242, p = 0.004) were associated with lower reported odds of the mastitis. Keeping heifers and cows in the same calving paddock was associated with higher estimated odds of mastitis, although the association was not statistically significant (OR = 2.46, p = 0.060) (Table 6). Some estimates were accompanied by relatively wide confidence intervals, reflecting limited observations within certain exposure categories and consequently reduced precision of the estimated effects. These extreme estimates likely reflect sparse data within certain exposure categories and should be interpreted with caution, with emphasis placed on the direction and consistency of associations rather than the magnitude of point estimates. Table 6Univariable mixed-effects logistic regression modeling for Block 6: mastitis management practicesVariableLevelOR (crude)SEzP> z95% CIDry cow therapy?No1.00Ref.Ref.Ref.Ref.Dried off cows1.1370.9640.150.8790.216 – 5.990Selected cows0.3830.31-1.190.2360.079 – 1.872Separate calving paddock?No1.00Ref.Ref.Ref.Ref.Yes1.40.9130.520.6060.390 – 5.027Heifers & cows same calving paddock?No1.00Ref.Ref.Ref.Ref.Yes2.4551.1711.880.060.964 – 6.255Mastitis main reason for culling?No1.00Ref.Ref.Ref.Ref.Yes2.3731.1111.850.0650.949 – 5.938Check udder health per cow?Daily before milking1.00Ref.Ref.Ref.Ref.Never1.2370.6840.380.7010.418 – 3.658Weekly1.7451.1250.860.3880.493 – 6.176When abnormal0.5480.552-0.60.550.076 – 3.945Record mastitis treatment?No1.00Ref.Ref.Ref.Ref.Yes0.9580.473-0.090.930.364 – 2.519Frequency of mastitis?Twice1.00Ref.Ref.Ref.Ref.Always0.3720.386-0.950.340.049 – 2.839More than thrice1.00(empty)*Thrice0.1770.213-1.440.150.017 – 1.873Who diagnosed mastitis?Farmer1.00Ref.Ref.Ref.Ref.LEO1(empty)*Other1.0781.2020.070.9460.121 – 9.578Veterinarian0.8890.573-0.180.8550.252 – 3.141How was mastitis diagnosed?Visual1.00Ref.Ref.Ref.Ref.Bacterial culture0.3330.203-1.810.0710.101 – 1.097CMT1.1521.4640.110.9110.095 – 13.905Treat mastitis?Yes1.00Ref.Ref.Ref.Ref.No0.2260.112-30.0030.086 – 0.597DrugOxytetracycline (L.A)1.00Ref.Ref.Ref.Ref.Gentamycin1(empty)*Herbs0.0950.144-1.560.1190.005 – 1.830Procaine-penicillin0.7620.901-0.230.8180.075 – 7.738Streptomycin1(empty)*Sulphonamides0.2860.362-0.990.3230.024 – 3.428Tramadol1(empty)*Trydox1(empty)*Tylosin1(empty)*PrescriberVeterinarian1.00Ref.Ref.Ref.Ref.Farmer0.980.74-0.030.9780.223 – 4.306 Univariable mixed-effects logistic regression modeling for Block 6: mastitis management practices Variables with p < 0.25 and biological plausibility were carried forward for multivariable analysis (Appendix Table 1). The summary of the decision for inclusion and exclusion is available in Appendix Table 1 and Supplementary material S3. Multivariable mixed-effects logistic regression Final models for each block, accounting for clustering at the LGA level, are presented in Table 7, and the key descriptions are presented below: In the adjusted model, farmer education level was not a significant predictor of mastitis (Table 7), and under the animal demographics and structure, the final model retained herd structure variables. Farms with few (1–3) female calves < 1 year had significantly lower odds of mastitis compared to those with a moderate number (4–6) (OR = 0.336, 95% CI: 0.125–0.904, p = 0.031) (Table 7). However, large herd size (based on total cattle) was a major risk factor. Farms with large herds had over six times the odds of mastitis compared to medium-sized herds (aOR = 6.557, 95% CI: 1.989–21.620, p = 0.002) (Table 7). While not statistically significant, notable associations included a markedly higher odds of mastitis associated with using well/borehole water (aOR = 19.757, p = 0.139) and a lower odds for providing feed supplements (aOR = 0.371, p = 0.235) (Table 7). Post-milking teat disinfection was strongly associated with lower odds (aOR = 0.013, 95% CI: 0.000–0.797, p = 0.039). Cleaning only dirty teats before milking also significantly reduced odds (aOR = 0.189, 95% CI: 0.065–0.550, p = 0.002). Conversely, using a separate cloth for cleaning teats was associated with higher odds (aOR = 4.921, 95% CI: 1.111–21.796, p = 0.036) (Table 7). Finally, not treating mastitis (aOR = 0.281, p = 0.059) and having no recurrent treatments (aOR = 0.283, 95% CI: 0.084–0.955, p = 0.042) were associated with lower odds. Sharing calving paddocks between heifers and cows significantly increased risk (aOR = 4.892, 95% CI: 1.026–23.322, p = 0.046) (Table 7). Table 7Outcomes of final multivariable mixed-effects logistic regression modelsVariableLevelOR (adjusted)SEZP > z95% CIFarmer and farm-level demographicsEducationIlliterate1.00Ref.Ref.Ref.Ref.College/Diploma0.9911.014-0.010.9930.133 – 7.368High school2.4822.1021.070.2830.472 – 13.047Primary school0.5040.274-1.260.2080.173 – 1.465University1----Animal demographics/structureFemale calves <1 yearModerate (4–6)1Ref.Ref.Ref.Ref.None (0)0.1470.274-1.030.3030.004 – 5.642Few (1–3)0.3360.17-2.160.0310.125 – 0.904Many (7–10)0.7040.334-0.740.460.277 – 1.786Very many (>10)0.7950.498-0.370.7140.233 – 2.710Heifers served awaiting confirmationLow (1–3)1Ref.Ref.Ref.Ref.None (0)1.6220.9490.830.4080.515 – 5.106Moderate (4–6)2.8671.6891.790.0740.904 – 9.095High (7–10)1.2320.9350.270.7840.278 – 5.451Very high (>10)0.3850.309-1.190.2340.080 – 1.853Housing and management practicesHerd size (total cattle)Medium herds1.00Ref.Ref.Ref.Ref.Small herds1.2160.7980.30.7650.336 - 4.399Large herds6.5573.9913.090.0021.989 - 21.620Working bullsFew (1–2)1.00Ref.Ref.Ref.Ref.Moderate (3–5)1.5790.9080.790.4270.511 - 4.874Many (6–10)0.2780.238-1.50.1340.052 - 1.486Very many (>10)0.3550.383-0.960.3370.043 - 2.938Herd size (total animals)Large herds1.00Ref.Ref.Ref.Ref.Small herds0.0540.098-1.590.1110.001 - 1.962Water and feed practicesWater sourceRiver/Stream1.00Ref.Ref.Ref.Ref.Lake/Dam0.1020.169-1.370.1690.004 - 2.642Other0.1170.221-1.140.2550.003 - 4.709Well/Borehole19.75739.8271.480.1390.380 - 1027.057Feed supplements?No1.00Ref.Ref.Ref.Ref.Yes0.3710.31-1.190.2350.077 - 1.88Milking hygiene practicesClean teats as preparation for milking?No1.00Ref.Ref.Ref.Ref.All cows0.2350.208-1.630.1020.041 - 1.335Only cows with dirty teats0.1890.103-3.050.0020.065 - 0.550Teat dip/spraying after milking?No1.00Ref.Ref.Ref.Ref.Yes0.013*0.028-2.070.0390.000 - 0.797Teat dip/spraying before milking?No1.00Ref.Ref.Ref.Ref.Yes3.9177.6900.70.4870.084 - 183.691Milking methodHand1.00Ref.Ref.Ref.Ref.Machine0.0030.013-1.420.1551.23E-06 - 8.664Use separate clothes for cleaning?No1.00Ref.Ref.Ref.Ref.Yes4.9213.7372.10.0361.111 - 21.796Mastitis managementTreat mastitis?Yes1.00Ref.Ref.Ref.Ref.No0.2810.189-1.890.0590.075 - 1.050Recurrence of mastitis treatment?Yes1.00Ref.Ref.Ref.Ref.No0.2830.176-2.030.0420.084 - 0.955Heifers & cows same calving paddock?No1.00Ref.Ref.Ref.Ref.Yes4.8923.8981.990.0461.026 - 23.322*Extremely small odds ratios may arise from sparse data within certain exposure categories and should be interpreted with caution Outcomes of final multivariable mixed-effects logistic regression models *Extremely small odds ratios may arise from sparse data within certain exposure categories and should be interpreted with caution Model diagnostics and fit Intra-class correlation coefficients from final mixed-effects logistic regression models The ICC estimates for each of the six conceptual blocks quantifies the proportion of total variance in the herd-level mastitis status (mastitis occurrence in the current year) attributable to clustering at the LGA level. This was highest in the water and feed block (ICC = 0.829, 95% CI: 0.376–0.975), indicating that a substantial proportion of the variability in mastitis risk associated with water and feed practices was due to differences across LGAs. Similarly, high ICC values were observed for the farmer and farm-level demographics (ICC = 0.767, 95% CI: 0.464–0.926) and the milking hygiene practices (ICC = 0.708, 95% CI: 0.442–0.881), suggesting strong contextual or geographic influences on these risk factors (Table 8). The animal demographics/structure and the housing and management practices also demonstrated moderately high ICC values (0.694 and 0.615 respectively), reflecting clustering at the LGA level but to a slightly lesser extent. The mastitis management had the lowest ICC of all models (0.581, 95% CI: 0.248–0.853) (Table 8), indicating that while management practices still varied by location, individual-level factors may play a more dominant role in shaping these practices. Table 8ICC estimates from final mixed-effects logistic regression models across all blocksModelLevelICCSE95% CIBlock 1: Farmer & Farm-Level DemographicsLGAs0.7670.1220.464 – 0.926Block 2: Animal Demographics/StructureLGAs0.6940.1190.432- 0.872Block 3: Housing & Management PracticesLGAs0.6150.1350.344 - 0.830Block 4: Water & FeedLGAs0.8290.1510.376 - 0.975Block 5: Milking Hygiene PracticesLGAs0.7080.1180.442 - 0.881Block 6: Mastitis ManagementLGAs0.5810.1780.248 - 0.853 ICC estimates from final mixed-effects logistic regression models across all blocks Model fit assessment using AIC and BIC statistics These information criteria provide a basis for comparing the relative goodness of fit of the models for the final mixed-effects logistic regression for the six thematic blocks, penalized for model complexity. The AIC and the BIC suggested that the water and feed (AIC = 131.210; BIC = 149.882) and the mastitis management (AIC = 133.530; BIC = 149.467) presented with the best model fit because they had the lowest AIC and BIC values (Table 9). The animal demographics/structure had the highest AIC (264.559) and BIC (301.463), indicating a poorer fit, the milking hygiene practices similarly had a relatively high AIC (228.767) and BIC (257.584), perhaps reflecting the model’s sensitivity to individual behavior and practice-level variations (Table 9). Table 9Model fit statistics (AIC and BIC) for final mixed-effects logistic regression models across all blocksModelObservationsLog Likelihood (Null)Log Likelihood (Model)DFAICBICBlock 1: Farmer & Farm-level Demographics231––86.0765182.151199.363Block 2: Animal Demographics/Structure296–-122.2810264.559301.463Block 3: Housing & Management Practices236–-91.3178198.633226.344Block 4: Water & Feed166–-59.6056131.21149.882Block 5: Milking Hygiene Practices271–-106.3848228.767257.584Block 6: Mastitis Management179–-61.7655133.530149.467DF Degree of freedom, AIC Akaike information criterion; BIC Bayesian information criterion Model fit statistics (AIC and BIC) for final mixed-effects logistic regression models across all blocks DF Degree of freedom, AIC Akaike information criterion; BIC Bayesian information criterion Discussion Summary of main findings This study assessed the prevalence and associated risk factors for clinical mastitis among lactating dairy cows in Plateau State, Nigeria, using a robust epidemiological and statistical framework. A hierarchical block-wise multivariable mixed-effects logistic regression approach was employed to evaluate the contribution of varied cow, farm, and management level factors for mastitis occurrence, while accounting for potential clustering at the LGA level. The overall prevalence of clinical mastitis in the current lactation cycle was 59.4%, consistent with findings reported in other regions of Nigeria [15, 17]. Our analyses revealed that the most significant predictors of mastitis were concentrated within three conceptual domains: mastitis management practices, milking hygiene, and aspects of housing and animal structure. Key risk factors included the failure to treat mastitis when identified, the absence of recurrence management protocols, and the co-housing of heifers and adult cows in the same calving paddock. The apparently lower odds of mastitis among farms reporting that they did not treat mastitis should be interpreted with caution. This counterintuitive result is unlikely to indicate that non-treatment reduces mastitis risk. Rather, it may reflect reverse causation, whereby farms experiencing more frequent, more severe, or more noticeable mastitis cases were more likely to administer treatment, while farms with fewer reported cases had no perceived need for treatment. It may also reflect differences in farmer recognition, treatment-seeking behaviour, or reporting practices. Consequently, treatment status in this setting may be better understood as an indicator of underlying disease occurrence than as an independent protective or harmful exposure. Residual confounding due to unmeasured factors such as disease severity, access to veterinary care, and farmer decision-making processes may also contribute to this observed association. These management lapses were associated with increased odds of mastitis, suggesting possible limitations in disease control awareness or constraints in veterinary service access. Milking hygiene practices, such as inadequate udder preparation, absence of post-milking teat dipping, and the use of shared or unclean towels, were also strongly associated with increased mastitis risk. For example, cows whose teats were not dipped after milking had significantly higher odds of developing mastitis, confirming that such practices are essential to prevent new intramammary infections during the post-milking period when teat sphincters remain dilated [6, 36–39]. This finding serves as a testament to the effect of hygiene in reducing mastitis risk under smallholder dairy production systems. The association observed for the use of separate cleaning cloths should be interpreted cautiously. Although the use of individual towels is generally recommended in mastitis control programs, improper washing, reuse of contaminated cloths, or inadequate drying between milking sessions may facilitate bacterial transmission between cows. In smallholder dairy systems, cloths used for udder cleaning are often reused across multiple cows without adequate washing or disinfection, which may inadvertently promote cross-contamination of mastitis pathogens. Previous studies have shown that contaminated udder cloths can serve as reservoirs for bacteria such as Staphylococcus aureus and Streptococcus spp., thereby increasing the risk of intramammary infection when hygiene protocols are not strictly followed. For example, studies in smallholder dairy systems in East Africa and Asia have documented that poor udder preparation practices, including reuse of inadequately cleaned cloths, can contribute to mastitis transmission during milking [6, 40]. Consequently, the observed association may reflect variations in cloth hygiene practices rather than the inherent effect of using separate cloths per se. The very small odds ratio observed for post-milking teat dipping should be interpreted with caution. Although the direction of the association is biologically plausible and consistent with established mastitis prevention practices, extremely small effect estimates may arise from sparse exposure categories or strong clustering effects within multilevel datasets. Therefore, interpretation should focus primarily on the direction of association and the confidence interval rather than the magnitude of the point estimate alone. Some predictors were associated with relatively wide confidence intervals, reflecting sparse observations within certain exposure categories and consequently reduced precision of the estimated effects. These estimates should therefore be interpreted with caution. In terms of structural and herd level variables, larger herds were associated with higher odds of mastitis, possibly reflecting factors such as increased animal density and management challenges [38, 41–44]. It should also be noted that herd size may act as a proxy indicator for broader management characteristics such as production intensity, labor availability, veterinary access, and biosecurity practices, which could partially confound the observed association with mastitis occurrence. Interestingly, farms with small herd sizes showed lower predicted probabilities of mastitis, suggesting that more individualized animal care may be associated with lower prevalence. The use of mixed-effects modeling allowed for the control of unobserved heterogeneity across LGAs, and the inclusion of ICCs further quantified the contribution of between-cluster variability to the model. The relatively high ICCs across the blocks underscored the importance of contextual and geographical factors in mastitis epidemiology. From a policy and intervention perspective, the observed clustering of mastitis risk across LGAs suggests that control strategies should not focus solely on individual animals but also on herd-level and community-level management practices. Programs aimed at improving mastitis control in smallholder dairy systems may therefore benefit from coordinated interventions at the local level, including farmer training on milking hygiene, strengthening veterinary extension services, and promoting routine mastitis monitoring practices. Such geographically targeted interventions may be more effective than isolated farm-level actions because farms within the same region often share similar environmental conditions, management systems, and access to veterinary resources. High ICC values indicate that a substantial proportion of the variation in mastitis occurrence was attributable to differences between clusters (LGAs) rather than individual farms alone, highlighting the importance of accounting for hierarchical data structures using multilevel modelling approaches [28]. This suggests that farms located within the same geographical area may share similar environmental conditions, management practices, veterinary service access, or disease exposure patterns that influence mastitis risk. The presence of such clustering justifies the use of multilevel mixed-effects modelling, as conventional single-level regression models would fail to account for this dependency structure and could underestimate standard errors and overstate statistical significance. The findings of this study reaffirm that mastitis remains a pervasive and economically significant disease among lactating dairy cows in Nigeria, consistent with prior research across multiple agro-ecological zones in the country. Our observed prevalence of clinical mastitis aligns with earlier studies that have documented both clinical and sub-clinical forms of the disease. Shittu et al. [15], for example, reported a sub-clinical mastitis (SCM) prevalence of 85.3% at the herd level and 43.3% at the quarter level among dairy cows in the Savannah region of Nigeria. Likewise, Zailani et al. [45] recorded a sub-clinical mastitis prevalence of 89.5% in Kanam Local Government Area, further supporting the endemicity of mastitis in northern and central Nigeria. The risk factors identified in our study including inadequate post-milking teat disinfection, poor udder hygiene, and the co-housing of heifers and adult cows are consistent with previous findings. In a study conducted by Ózsvári & Ivanyos [46], failure to practice teat dipping and use of contaminated milking equipment were associated with elevated mastitis risk. Similarly, Junaidu et al. [47] and Shittu et al. [15] highlighted poor milking hygiene and lack of veterinary input as major contributors to mastitis prevalence in smallholder dairy farms in Sokoto State. Globally, our findings resonate with similar studies in low- and middle-income countries (LMICs), where mastitis prevalence and risk factors tend to be shaped by suboptimal hygiene, limited veterinary services, and traditional management systems. In Dodoma, Tanzania, Mramba et al. [5] found a prevalence of mastitis at cow level as 59.8%, attributing the high burden to cleanliness (hand and udder washing before milking). Nyokabi et al. [40] also reported a strong link between poor milking hygiene and mastitis among smallholder dairy farmers in central Kenya, emphasizing that preventive measures such as hand washing, udder cleaning, and dry cow therapy are often overlooked. Comparable associations have been observed outside Africa. In India, Sharma et al. [48] documented a mastitis prevalence of 31% in Deoni cows and 65% in crossbred cows, with risk factors including stage of lactation, milk yield, parity, and mastitis treatment history. In Brazil, a study by Tomazi et al. [49] highlighted that housing system and bulk milk somatic cell count were significant herd-level risk factors associated with clinical mastitis, aligning with findings from our study on the influence of housing and udder health management practices. These parallels illustrate that while prevalence rates may vary across regions, the fundamental risk factors remain remarkably similar, particularly in settings where veterinary infrastructure and farmer training are limited. Additionally, our findings corroborate those from industrialized countries where interventions such as post-milking teat dipping, use of individual towels, and dry cow therapy have substantially reduced mastitis prevalence [6, 39]. The contrast in mastitis control success between high-income and low-resource settings underscores the need for scalable, farmer-centric interventions that align with local socio-economic realities. This study contributes to the growing body of evidence indicating that mastitis occurrence in Nigeria and much of sub-Saharan Africa is associated with several modifiable management and hygiene-related factors. Addressing these through targeted hygiene practices, structured training programs, and improved veterinary extension services could yield substantial improvements in dairy productivity and animal welfare. While several associations were identified in this study, the results should be interpreted cautiously within the context of observational cross-sectional data. Some associations may reflect underlying management patterns, reporting behaviors, or contextual factors rather than direct causal mechanisms. For example, differences in farmer knowledge, access to veterinary services, herd management practices, and environmental conditions across LGAs may partly explain the observed variation in mastitis occurrence. Additionally, some predictors may act as proxies for broader management systems rather than independent determinants of disease risk. Considering these alternative explanations is important when interpreting the magnitude and direction of the associations identified in this study. Strengths of the study This study has several methodological and analytical strengths that enhance the robustness and credibility of the findings. First, the adoption of a hierarchical block-wise modeling strategy allowed explanatory variables to be organized into conceptually related domains ranging from farmer characteristics to mastitis management practices. This structured modelling approach helped reduce overfitting, minimized multicollinearity, and improved the interpretability of risk factor relationships [28, 50]. Second, the use of multivariable mixed-effects logistic regression with random intercepts at the LGA level accounted for clustering within the hierarchical data structure. This approach improves the reliability of statistical inference by addressing intra-cluster correlation, which is common in epidemiological studies involving farms nested within geographic regions [51, 52]. Third, rigorous data management and validation procedures were implemented, including logical verification of responses, harmonization of categorical variables, and careful handling of missing values. These steps improved data consistency and strengthened the reliability of the statistical analyses [35]. Finally, the study incorporated comprehensive model diagnostics and post-estimation analyses, including residual evaluation and marginal effects visualization [34, 53, 54]. These procedures enhanced model transparency and facilitated clearer interpretation of the predicted probabilities of mastitis occurrence across different risk factors. Limitations Despite the methodological rigor and analytical strengths of this study, several limitations should be acknowledged to contextualize the findings and guide future research. Firstly, the cross-sectional study design limits the ability to infer causal relationships between identified risk factors and the occurrence of clinical mastitis. Although associations were established, the temporal direction of these relationships remains unclear. For example, it is uncertain whether certain management practices preceded or followed the onset of mastitis. Longitudinal or prospective cohort studies are better suited to establishing causal links and understanding the dynamics of disease progression over time [55, 56]. Secondly, the study relied on self-reported information for key explanatory variables, including milking hygiene practices, mastitis treatment history, and feed supplementation. While we acknowledged that self-reported data are inherently vulnerable to recall bias, especially when farmers are required to remember routine practices or past events over an extended period, the reliance on farmer-reported mastitis in this study reflects field realities in low-resource smallholder settings where diagnostic facilities such as CMT or somatic cell counting may not be available always, which, if carefully implemented, may produce credible outcomes [22–24, 57]. In addition, because mastitis cases were identified based on farmer self-reports rather than laboratory confirmation, some degree of reporting bias may have occurred if farmers differed in their ability to recognize or recall mastitis signs. Additionally, social desirability bias may have led respondents to overreport good practices such as hand washing or teat disinfection, which could attenuate observed associations with mastitis risk [58]. Triangulating questionnaire responses with direct observations or farm records could enhance data validity in future studies. Thirdly, the study focused exclusively on clinical mastitis, defined by the presence of visible signs of udder inflammation or abnormal milk. However, SCM – characterized by elevated SCC in the absence of overt symptoms is typically more prevalent and economically significant, particularly in smallholder settings where detection is challenging [15, 59]. By excluding SCM, the study may have underestimated the true burden of mastitis and missed important risk factors uniquely associated with sub-clinical presentations. Another limitation is the lack of explicit modeling of clustering at the farm or household level. While random intercepts at the LGA level were included to account for geographic clustering, individual farms or households represent an additional layer of dependency. Although mastitis status was analysed at the herd level, additional within-farm clustering of individual cow-level outcomes could not be explicitly modelled because the available data were aggregated at farm level. Ignoring such clustering may lead to residual confounding and underestimated standard errors [60]. Future research should consider multi-level modeling with additional random effects at the farm level to more accurately capture data structure. Finally, sampling bias may also be present. The selection of LGAs was purposive, based on the presence of active dairy farming, and farm recruitment was facilitated through extension officers. While this ensured access to relevant participants, it may have inadvertently excluded harder-to-reach or less engaged farmers, limiting the generalizability of findings to the wider dairy population in Plateau State and beyond [61]. However, in this work, we utilised the purposive selection of LGAs to ensure representation across diverse agro-ecological and production systems (intensive, semi-intensive, and extensive), consistent with FAO and ILRI livestock surveillance frameworks [62]. This stratified purposive design enhances ecological validity rather than compromising randomness, and with approximately 300 herds across ten LGAs, post-hoc power analysis indicates sufficient power (> 0.8) to detect moderate effect sizes (OR ≥ 2.0) at 95% confidence. While these limitations do not invalidate the study’s findings, they highlight the need for cautious interpretation and underscore opportunities for methodological enhancement in future investigations. Implications for policy and practice Recommendations for mastitis control and prevention From a practical perspective, the findings highlight the need for targeted mastitis control strategies in smallholder dairy systems. Priority interventions should include strengthening farmer training on proper milking hygiene and udder health management, promoting routine mastitis monitoring using simple on-farm diagnostic tools such as the CMT, and improving access to veterinary advisory services. In addition, herd-level management practices, particularly those related to housing hygiene and mastitis record-keeping, should be emphasized within extension programs to support early detection and prevention of mastitis. This study underscore the urgent need to develop and implement comprehensive mastitis control and prevention strategies tailored to the realities of smallholder and semi-commercial dairy farms in Plateau State and similar settings in sub-Saharan Africa. The significant associations found between poor milking hygiene, inadequate mastitis management practices, and increased mastitis risk highlight key areas for intervention. Improving milking hygiene practices should be a top priority. This includes hand washing before and after milking, using clean and preferably disposable or dedicated udder cleaning cloths for each cow, and practicing proper pre and post milking teat disinfection using effective antiseptics such as iodine based or chlorhexidine-based teat dips. Several studies have shown that consistent implementation of these practices significantly reduces the incidence of both clinical and subclinical mastitis [15, 46, 63–65]. Additionally, the study found a significant relationship between housing design and mastitis occurrence. Therefore, promoting the adoption of housing systems that enhance cleanliness, ventilation, and drainage, while reducing overcrowding and standing time on hard surfaces, could contribute to mastitis prevention. Well-designed housing systems reduce the accumulation of fecal material, mud, and urine, which are key reservoirs of environmental pathogens like Escherichia coli and Streptococcus uberis [6, 66–68]. Strengthening mastitis monitoring systems through farmer training, early recognition of clinical signs, and the adoption of affordable on-farm diagnostic approaches – such as the California Mastitis Test (CMT) where feasible – could substantially improve early detection and control of mastitis in smallholder dairy systems [69–72]. Establishing mastitis control protocols at the farm level clearly outlining steps for prevention, detection, isolation, and treatment will also help standardize responses and improve outcomes. These protocols should be participatorily developed in collaboration with veterinary experts, community animal health workers and the farmers and adapted to the local context for sustainability. Farmer training and veterinary outreach Achieving meaningful reductions in mastitis incidence requires sustained capacity building for farmers and their supporting networks. The study highlights knowledge gaps in proper milking hygiene and mastitis treatment practices, suggesting the need for regular and targeted training programs. Such training should focus on the causes and consequences of mastitis, proper milking routines, animal husbandry, and early signs of disease. Studies have shown that education interventions significantly improve farmer compliance with best practices and result in lower SCC and mastitis prevalence [73–77]. In addition to training, enhanced veterinary outreach and extension services are vital. Mobile veterinary clinics, availability of qualified community animal health workers, periodic LGA-based herd health days, and village-level demonstration farms would help sustain farmers’ continued herd health education, while bringing needed expertise closer to remote farmers. These services should not only provide clinical care but also serve as platforms for disease monitoring and reporting mandate of each LGA and the State, farmer education, and data collection. Furthermore, policies that promote affordable access to veterinary care via rural-based qualified community animal health workers (CAHWs), will be instrumental in sustaining control efforts for remote farmers. Government subsidies or public-private partnerships could facilitate the availability of all required resources for efficient herd health management in underserved areas. Encouraging the formation of farmer cooperatives or associations may also increase bargaining power for bulk purchases and enhance peer learning. Finally, integration of mastitis prevention into broader animal health and food safety policies will help align mastitis control efforts with national livestock development goals. This includes linking mastitis surveillance with milk quality monitoring and antimicrobial resistance (AMR) strategies given the role of mastitis in driving antibiotic use on farms [78]. Conclusion This study provides important insights into the epidemiology of clinical mastitis in dairy cows in Plateau State, Nigeria. Using a hierarchical multivariable mixed-effects logistic regression framework, several management and herd-level factors were found to be associated with mastitis occurrence. In particular, inadequate post-milking teat disinfection, poor udder hygiene practices, larger herd sizes, and weak mastitis record-keeping practices were associated with higher reported mastitis occurrence. The substantial clustering observed at the LGA level further highlights the influence of shared environmental and management contexts on mastitis risk. These findings highlight the importance of strengthening mastitis prevention and control strategies in smallholder and peri-urban dairy systems. Priority actions should include improving farmer awareness of udder hygiene practices, strengthening routine mastitis monitoring and record-keeping, and expanding access to veterinary advisory services. Practical interventions such as improved milking hygiene and appropriate mastitis detection practices could contribute to reducing the disease burden under field conditions. Future studies should consider longitudinal designs and incorporate diagnostic confirmation of mastitis cases to better understand causal relationships and the effectiveness of specific interventions. Overall, improving mastitis control practices has the potential to enhance dairy cow health, milk quality, and the economic sustainability of dairy farming households in Plateau State and similar production systems across sub-Saharan Africa. While these findings provide important insights into potential determinants of mastitis in Nigerian dairy systems, the observational nature of the study limits causal inference. The identified associations should therefore be interpreted cautiously and may serve to guide future longitudinal and intervention studies. Nevertheless, promoting improved milking hygiene and herd-level management practices remains a plausible and evidence-informed strategy for reducing mastitis burden. Supplementary Information Supplementary Material 1. Acknowledgements Mrs C Vance, Dr J Suarez Martinez and Dr M Christian for assistance with planning and design; Dr. Usman Adamu Rayyanu for his assistance with the field work and the farmers for their availability and cooperation. The authors gratefully acknowledge the support and collaboration of the livestock extension officers who facilitated access to farms across various LGAs. We are deeply thankful to the dairy farmers and households who generously participated in the study and shared their time, experiences, and knowledge. We also acknowledge the valuable technical and logistical assistance provided by field enumerators, data entry clerks, and veterinary personnel involved in the collection and processing of field data. Special thanks go to the management and staff of the National Veterinary Research Institute, Vom, Nigeria, for providing institutional support and ethical oversight throughout the study.Finally, we appreciate the contributions of colleagues and mentors who provided feedback during the development and review of the study protocol and manuscript. Abbreviations Animal Ethics Committee Akaike’s Information Criterion Antimicrobial resistance Adjusted Odds Ratio Bayesian Information Criterion Community animal health workers California Mastitis Test Directed Acyclic Graph Food and Agriculture Organization of the United Nations Intraclass correlation coefficient International Livestock Research Institute Local government area Low- and middle-income countries Likelihood-ratio test National Veterinary Research Institute, Nigeria Somatic cell count Sub-clinical mastitis Supporting Evidence-Based Intervention Odds Ratio Confidence interval Sub-Saharan Africa Variance Inflation Factors Authors’ contributions Conceptualization: JO, M-BB and AP; Data curation and field investigation: M-BB, JO, AS and FOF; Methodology, formal analysis and visualization: AS and FOF; Project administration: and supervision: AP, JO, and M-BB; Software: AS and FOF; Resources: AP, JO, AS and FOF; Writing – original draft: AS, JO and FOF; Writing – review & editing of the approved draft: All authors. Funding This field work was funded by Gates Foundation Grant Number INV-004102. Data availability All data generated or analysed during this study are included in this published article. Declarations Ethics approval and consent to participate The questionnaire tool was approved by Supporting Evidence-Based Intervention (SEBI) of the Royal (Dick) School of Veterinary Studies, University of Edinburgh. The study was conducted in accordance with the local legislation and institutional requirements of the NVRI, Vom, Nigeria and ethical clearance was obtained from the NVRI Ethics Board (Approval number: AEC/03/210/25). Informed Consent was sought and received from all participants. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated or analysed during this study are included in this published article.