during blood donation screening. Anthropometric and hematological data were extracted from electronic health

records. The relationship between weight (kg) and Hb (g/dL) was analyzed using robust statistical methodologies

in Python, including Spearman’s correlation and multivariate linear regression, controlling for sex, age, and ART

duration. A machine learning framework (Random Forest) was implemented for predictive validation.

Analysis revealed a powerful positive correlation between weight and Hb (ρ = 0.82, *p* < 0.001). Multivariable

regression confirmed weight as a highly significant independent predictor, with every 10 kg increase associated

with a +1.2 g/dL gain in Hb (β = 0.12, 95% CI: 0.08–0.16, *p* < 0.001). A profound sex-based disparity was

identified, with female sex independently associated with a -1.45 g/dL Hb deficit (*p* = 0.001). The machine

learning model validated weight as the paramount predictive feature (importance score = 0.89), forming the basis

Keywords: HIV anemia, nutritional status, body weight, machine learning, predictive modeling, sub-Saharan

Africa, health disparities

Despite global success in scaling up antiretroviral therapy (ART), 20–40% of persons living with HIV (PLWH)

in sub-Saharan Africa continue to experience wasting a devastating complication that perpetuates anemia and

undermines treatment efficacy(1&2). This synergy creates a vicious cycle of declining health that existing, siloed

clinical guidelines fail to address(3&4). Can we leverage the simple, routinely measured metric of body weight

to predict and prevent the onset of severe anemia, transforming HIV care from reactive management to proactive,

precision prevention?(5&6)

(Hb) dynamics, are not fully restored, perpetuating significant morbidity and functional impairment (12).

The pathophysiology of anemia in HIV is complex and multifactorial, primarily driven by: (1) Chronic immune

activation leading to elevated hepcidin and subsequent iron sequestration (13); (2) Nutritional deficiencies of

iron, vitamin B12, and folate, often exacerbated by food insecurity (14); and (3) ART-related side effects,

including zidovudine-induced bone marrow suppression (15). Despite clear clinical awareness of this synergy,

no standardized, integrated interventions exist specifically targeting the wasting-anemia complex(16&17).

Current WHO guidelines focus predominantly on isolated anemia correction through iron supplementation or

ART switches, rather than on holistic weight-Hb optimization, leaving a critical gap in comprehensive patient

management(18&19).

Lack of Quantitative Dynamic Models: Prior studies rely on categorical classifications (e.g.,

underweight/normal) rather than utilizing continuous weight metrics to model the precise, dynamic correlation

with Hb levels over time(25).

Absence of Sex-Specific Risk Stratification: Current clinical algorithms treat anemia risk as uniform across

sexes, ignoring pronounced physiological and sociological disparities that demand precision public

health approaches(26&27).

No Predictive Capacity: The field lacks validated tools to forecast individual Hb trajectories based on early

weight changes, representing a missed opportunity for preemptive intervention before clinical anemia

manifests(28&29).

hemoglobin levels, independent of ART duration, but this association is significantly modified by sex, being

stronger in women due to their heightened baseline anemia susceptibility and different iron homeostasis.

determinants (food security, gender norms) to explain health outcomes(30). Furthermore, we employ Complex

Systems Theory to understand the non-linear, interacting dynamics between weight, inflammation, and

hematopoiesis(31&32).

This research aims to bridge the translation gap between epidemiological observation and clinical action through

four specific objectives:

Primary: To precisely quantify the continuous weight-Hb correlation using advanced longitudinal modeling

(linear mixed-effects regression) in a cohort of 500 HIV-positive adults.

Secondary: To identify sex-specific risk thresholds where weight loss predicts clinically significant anemia,

enabling targeted interventions.

Methodological: To develop and validate an interpretable machine learning framework (XGBoost combined

with SHAP analysis) to predict individual Hb decline from early weight loss patterns(33).

guidelines(34). It offers a novel, integrated framework for HIV care that is proactive, personalized, and feasible

for the resource-constrained settings that bear the greatest burden of the HIV epidemic(35&36).

Ultimately, this research seeks to provide clinicians with a practical, evidence-based strategy to use a simple,

existing metric body weight to preempt a serious complication, thereby improving the quality of life and long-

term outcomes for millions of people living with HIV globally(37&38).

A retrospective cohort design was employed to investigate the longitudinal relationship between body weight

and hemoglobin (Hb) levels in a population of HIV-positive adults on antiretroviral therapy (ART)(39&40).

This quantitative, observational approach was selected as the optimal methodology for several reasons. First, it

The target population consisted of 1,200 HIV-positive adults receiving ART at six tertiary hospitals in Nigeria

between 2019 and 2023(43&44). To ensure the sample was representative and minimized selection bias, a

stratified random sampling technique was utilized(45&46). The stratification was based on three key variables

known to influence the outcomes: sex (50% male, 50% female), BMI categories (underweight, normal,

overweight), and ART regimen class (NNRTI-, PI-, and INSTI-based therapies)(47). This ensured that the

analytical sample reflected the diversity of the broader clinical population.

Inclusion Criteria:

Adults aged ≥18 years with a confirmed HIV-1 infection. This ensured biological maturity and a homogeneous

disease pathology(48).

Documented weight and Hb measurements at baseline and at a minimum of two follow-up visits. This was

crucial for establishing a trend over time(49).

Continuous ART adherence for ≥6 months prior to enrollment. This criterion was essential to isolate the effects

of a stable therapeutic regimen and avoid confounding from recent treatment initiation or poor adherence(50).

A history of blood transfusion within three months prior to data extraction. Transfusions artificially elevate Hb

levels, providing a misleading picture of the patient's true hematological status(53).

After applying these criteria, the final analyzed sample comprised 857 participants.

Data collection involved the extraction of pre-existing clinical measurements from Electronic Health

Records( EHRs) and laboratory systems(54).

Anthropometric Measurements:

Weight: Was measured during clinical visits using calibrated SECA 803 digital scales, which have a precision

of ±0.1 kg. Measurements were taken with participants wearing light clothing to ensure consistency(55).

Height: Was assessed using a wall-mounted stadiometer to calculate Body Mass Index (BMI; kg/m²) (56).

markers, specifically C-reactive protein (CRP) and serum ferritin, were measured using enzyme-linked

immunosorbent assay (ELISA) kits from stored serum samples.(59)

demographic and clinical parameters, summarized in Table 1, established the profile of the study population.

The median duration on antiretroviral therapy (ART) was 22 months (IQR: 12–34), indicating a population with

Table 1. Baseline Demographics and Clinical Parameters

The solid line represents the regression line, showing the predicted hemoglobin value for any given weight. The

two dashed lines represent the 95% confidence interval, indicating the range within which we are 95% confident

the true regression line lies. The strong upward slope and narrow confidence band visually confirm the powerful,

significant positive correlation described statistically by Spearman's ρ.

a Random Forest machine learning model was deployed (n_estimators=200). The feature importance analysis,

0.72). This independent, non-parametric method confirmed the paramount importance of weight, as identified

in the regression analysis.

(A horizontal bar chart where the length of each bar represents the mean absolute SHAP value (feature

(95% CI: 0.78–0.90).

To visualize the clinical impact of weight change over time, participants were stratified into three mutually

exclusive categories based on their net weight change over the study period: Weight Gain (≥5 kg increase,

n=15), Weight Stable (change within ±2 kg, n=12), and Weight Loss (≥5 kg decrease, n=11). The trajectories of

their mean hemoglobin levels over the 18-month follow-up are depicted in Figure 3.

The data revealed starkly divergent paths. The Weight Gain group experienced a substantial and sustained

increase in hemoglobin, with a mean rise of +1.8 g/dL from baseline to 18 months (95% CI: 1.2–2.4, p < 0.001).

Conversely, the Weight Loss group exhibited a profound and statistically significant decline in hemoglobin, with

a mean decrease of -2.1 g/dL over the same period (95% CI: -2.7 to -1.5, p < 0.001). The Weight Stable group

maintained relatively constant hemoglobin levels throughout the follow-up, with minimal fluctuation.

Fig. 3. Hemoglobin Trajectory Over Time by Weight Change Category.

*(A line graph with time (months) on the x-axis and mean hemoglobin (g/dL) on the y-axis. Three distinct lines

are plotted: a green upward-sloping line for "Weight Gain ≥5kg", a blue flat line for "Weight Stable (±2kg)",

and a red downward-sloping line for "Weight Loss ≥5kg". Shaded areas around each line represent the 95%

confidence intervals. The graph is titled "Longitudinal Change in Hemoglobin by Weight Change Category".)*

viral suppression through ART(70). The results compel a re-evaluation of current clinical monitoring practices

and suggest that simple anthropometric measurements could serve as vital, low-cost tools in the management of

HIV-related comorbidities(71).

Three major findings emerge from our analysis, each carrying significant implications for both clinical practice

and research.

Our analysis revealed a remarkably strong positive correlation between weight and Hb (Spearman’s ρ =

0.82, *p* < 0.001). The multiple linear regression model further quantified this relationship, demonstrating that

every 10 kg increase in weight was associated with a +1.2 g/dL gain in Hb, even after adjusting for sex, age, and

ART duration. This finding aligns with the physiological understanding that nutritional status is a cornerstone

of erythropoiesis(72). Weight integrates multiple pathways: it is a proxy for adequate caloric and protein intake

metabolism are not yet fully understood(76). The fact that our model explained 58% of the variance in Hb levels

(R² = 0.58) is striking; it suggests that weight is not merely a correlate but a central hub that encapsulates a large

portion of the multifactorial etiology of anemia in this population(77).

A critical and sobering finding was that female sex was independently associated with a 1.45 g/dL lower Hb

level compared to males (*p* = 0.001)(78). This disparity is consistent with reports from other African cohorts,

such as in Ethiopia, who found differences of 0.8–1.2 g/dL. However, it contrasts with data from high-income

countries like the United States, where studies suggest that sex-based hematologic differences diminish

significantly with sustained ART. This divergence is highly instructive. It implies that while ART is necessary,

it is not sufficient to overcome the entrenched biological and sociocultural factors that drive health inequities in

inequities where women may have less access to nutrient-rich foods, higher rates of parasitic infections, and

Our study advances the field by moving beyond traditional statistics to implement a predictive machine learning

80. Munro, M. G., Mast, A. E., Powers, J. M., Kouides, P. A., O’Brien, S. H., Richards, T., ... & Levy, B.

81. Maina Mwaura, F. Interplay Between Infant Nutrition and Malaria Susceptibility in West Africa:

82. Rodriguez-Diaz, C. E., Davis, W., Ellis, M. V., Cameron, M. S., Donastorg, Y., Bowleg, L., ... &

83. Daif, N., Di Nunno, F., Granata, F., Difi, S., Kisi, O., Heddam, S., ... & Zounemat-Kermani, M. (2025).

84. Uyoga, S., George, E. C., Bates, I., Olupot‐Olupot, P., Chimalizeni, Y., Molyneux, E. M., & Maitland,