Spatial patterns and determinants of Anemia in women of reproductive age in Zambia (2018-2024): A multilevel ordinal regression approach

This study analyzes spatial patterns and determinants of anemia severity among 19,362 Zambian women of reproductive age from 2018 to 2024, revealing that HIV status, pregnancy, and rural residency significantly increase risk while union status and financial access are protective, alongside a non-random geographic distribution with expanding hotspots in the Western, North-Western, and Luapula regions.

The Gist

Imagine the human body as a bustling city. In this city, hemoglobin (a protein in your blood) acts like the delivery trucks that carry oxygen to every neighborhood. When there aren't enough trucks, the city's lights dim, the factories slow down, and the people feel tired, weak, and unable to think clearly. This condition is called anemia.

This research paper is like a detective story investigating why this "oxygen shortage" is happening to women of childbearing age in Zambia, and where the worst "blackouts" are occurring. The researchers looked at data from two different years, 2018 and 2024, to see if things were getting better, worse, or just moving around.

Here is the story of their findings, broken down into simple parts:

1. The Big Picture: A Stagnant Problem

The researchers checked the "city lights" in 2018 and again in 2024. They found that the problem hasn't really gone away.

• The Stat: About 31% of women were anemic in 2018, and 30% in 2024.
• The Analogy: Imagine trying to fix a leaky roof. You might patch a few holes, but if the rain keeps coming in at the same rate, the house stays just as wet. The overall number of women with anemia hasn't dropped significantly over six years. It remains a "moderate public health emergency."

2. The Culprits: Who is Most at Risk?

The study used a special kind of math (like a multi-layered filter) to figure out what makes the "delivery trucks" stop working. They looked at three layers: the individual, the household, and the community.

The Individual Layer (The Person):

• HIV Status: Being HIV-positive is like having a broken engine in your delivery truck. It makes anemia 2.6 times more likely. The virus and the medicines needed to fight it can interfere with blood production.
• Pregnancy: Pregnancy is like a construction project that needs double the resources. A pregnant woman's body needs way more iron to build a baby. If she doesn't get extra fuel, her own "oxygen trucks" run dry. Pregnant women were twice as likely to have severe anemia.
• Age: Younger women (teens) and older women (40+) were more at risk, similar to how a new car or an old car might have more mechanical issues than a mid-life vehicle.

The Household Layer (The Family):

• Money for Medical Help: If a woman says, "I can't afford to see a doctor," she is at higher risk. It's like having a flat tire but no money for a repair shop. The truck stays broken.
• Marital Status: Interestingly, being married was a "shield." Married women had lower odds of severe anemia, perhaps because they have more support or better access to resources.

The Community Layer (The Neighborhood):

• Rural vs. Urban: Living in the countryside (rural areas) was a risk factor. Think of rural areas as remote villages where the "supply trucks" (doctors, iron supplements, diverse food) have a harder time getting there compared to the busy city centers.
• Location: Where you live matters immensely. Women in certain provinces (like Western, North-Western, and Luapula) were living in "danger zones."

3. The Map: Where are the "Blackout Zones"?

This is where the study gets really cool. The researchers didn't just look at numbers; they looked at a map. They used a technique called Spatial Analysis, which is like using a thermal camera to see where the heat (anemia) is concentrated.

• The Hotspots: In 2018, the "heat" was mostly in the Western province. By 2024, the heat had spread northward. The North-Western and Luapula provinces became new hotspots.
• The Analogy: Imagine a game of "Whac-A-Mole." You hit the mole in the Western province, but by 2024, the moles popped up in the North-West and Luapula. The problem didn't disappear; it just moved.
• The Clusters: The study found that anemia isn't random. It clusters together like a storm system. There is a massive "storm front" in the south-west of Zambia (covering Western, Southern, and North-Western provinces) that has been there consistently for years.

4. Why is this happening?

The paper suggests that while we know who is sick, the why is a mix of things:

• Malaria: These hotspot areas are also where malaria is common. Malaria is like a thief that steals your red blood cells.
• Poverty: You can't buy iron-rich food if you have no money.
• Healthcare Access: If the nearest clinic is a day's walk away, you won't get your iron supplements.

5. The Takeaway: What Should We Do?

The authors conclude that we can't just use a "one-size-fits-all" approach for the whole country.

• The Old Way: "Let's give iron pills to everyone in Zambia."
• The New Way (Recommended): "Let's send a specialized rescue team to the Western, North-Western, and Luapula provinces."

They suggest that interventions need to be geographically targeted. It's like putting a fire extinguisher right next to the kitchen where the fire started, rather than just handing out fire extinguishers to the whole neighborhood.

In Summary:
Anemia in Zambian women is a stubborn, moving target. It is fueled by HIV, pregnancy, poverty, and malaria. While the overall numbers haven't changed much, the location of the worst cases has shifted north. To fix this, we need to stop treating the whole country the same and start targeting the specific "blackout zones" with better nutrition, malaria control, and healthcare access.

Technical Summary

1. Problem Statement

Anemia remains a critical public health challenge in low-and-middle-income countries (LMICs), particularly affecting women of reproductive age (WRA). In Zambia, anemia prevalence has historically been high (>30%), yet previous studies have often treated anemia as a binary outcome (anemic vs. non-anemic), masking variations in severity. Furthermore, there is a lack of research integrating geospatial analysis with multilevel modeling to understand how individual, household, and community factors interact to drive anemia severity over time. This study addresses these gaps by assessing the spatial patterns and determinants of anemia severity across two survey cycles (2018 and 2024) to inform targeted subnational interventions.

2. Methodology

Data Source and Study Design:

• Data: Secondary data from two waves of the Zambia Demographic and Health Survey (ZDHS): 2018 ($n=12,735$) and 2024 ($n=6,627$), totaling 19,362 women aged 15–49.
• Design: Analytical cross-sectional study.
• Outcome Variable: Anemia severity, categorized as an ordered categorical variable based on WHO hemoglobin (Hb) cut-offs:
  • Non-anemic (Hb $\ge$ 120 g/L for non-pregnant, $\ge$ 110 g/L for pregnant).
  • Mild, Moderate, and Severe anemia.
• Predictors: Variables were categorized into three levels:
  • Individual: Age, HIV status, pregnancy, marital status, employment, parity, etc.
  • Household: Wealth index, financial barriers to medical care, family size, sex of household head.
  • Community: Residency (urban/rural), Province.

Statistical Analysis:

1. Multilevel Ordinal Logistic Regression:

   • A three-level hierarchical model (Individual nested in Household nested in Community) was employed to account for data dependency.
   • Model Selection: Five random-intercept models were tested (Null, Individual-only, Household-only, Community-only, and Full Model). The best-fit model (Model RI5) was selected based on Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC).
   • Metrics: Adjusted Odds Ratios (AOR), 95% Confidence Intervals (CI), Variance Partition Coefficients (VPC), and Intra-class Correlation Coefficients (ICC) were calculated.
   • Assumption Check: The Brant test was used to check the proportional odds assumption; despite violations for age, births, and residency, the model was retained for parsimony and interpretability.

2. Spatial Analysis:

   • Global Autocorrelation: Global Moran's $I$ was used to assess overall spatial clustering.
   • Local Hotspots: Local Moran's $I$ and Getis-Ord $Gi^*$ statistics identified specific high-risk (hotspots) and low-risk (cold spots) clusters.
   • Cluster Detection: Spatial scan statistics (Kulldorff's SatScan) using a Bernoulli model identified primary and secondary clusters of anemia risk.
   • Software: Analyses were conducted in R (v5.2.1) and ArcGIS.

3. Key Contributions

• Methodological Advancement: Unlike previous studies that dichotomized anemia, this study utilizes ordinal regression to model severity levels, providing a more nuanced understanding of the disease burden.
• Multilevel Insight: It quantifies the proportion of variance attributable to unobserved household and community factors (VPC), demonstrating that a significant portion of anemia risk is contextual rather than purely individual.
• Temporal-Spatial Dynamics: By comparing 2018 and 2024 data, the study reveals the evolution of spatial hotspots, identifying a northward shift in high-risk areas that static analyses would miss.
• Policy Relevance: The findings move beyond national averages to identify specific provinces and districts requiring urgent, geographically targeted interventions.

4. Key Results

Prevalence:

• Anemia prevalence remained stable and high: 31.0% in 2018 and 30.4% in 2024 ($p=0.559$).
• Severity breakdown: Mild (16.3%), Moderate (13.2%), Severe (1.4%).

Determinants of Anemia Severity (Multilevel Model):

• Individual Level:
  • HIV Positive: Strongest risk factor (AOR = 2.63; 95% CI: 2.25–3.09).
  • Pregnancy: Significantly increased odds (AOR = 1.96; 95% CI: 1.67–2.31).
  • Marital Status: Being "currently in union" was protective (AOR = 0.66) compared to never being in a union.
  • Age: Non-linear relationship; women aged 20–29 had lower odds than adolescents (15–19), while older women (40–44) showed slightly higher odds.
• Household Level:
  • Financial Barriers: Not having money as a "big problem" for medical help was protective (AOR = 0.85).
• Community Level:
  • Residency: Rural residence was associated with higher severity (AOR = 1.21).
  • Geography: Significant provincial variations existed. Compared to the Central province, Western (AOR = 2.10), North-Western (AOR = 2.00), and Lusaka (AOR = 1.80) had the highest odds.

Spatial Patterns:

• Clustering: Anemia was non-randomly distributed. Global Moran's $I$ was significant for both years (2018: $I=0.147$; 2024: $I=0.130$; $p<0.001$).
• Hotspot Evolution:
  • 2018: Hotspots were concentrated in the Western province.
  • 2024: Hotspots expanded northward to include North-Western and Luapula provinces.
• Primary Clusters: Spatial scan statistics identified a persistent primary cluster in the south-west region (Western, Southern, and North-Western provinces) for both waves.
  • 2018 Primary Cluster: RR = 1.23 ($p<0.001$).
  • 2024 Primary Cluster: RR = 1.29 ($p<0.001$).

5. Significance and Conclusion

The study concludes that anemia in Zambian WRA is a persistent, moderate public health problem driven by complex interactions across individual, household, and community levels. While overall prevalence has not significantly decreased, the geographic distribution is dynamic, with high-risk areas shifting from the Western province to the North-Western and Luapula regions.

Implications for Policy:

• Targeted Interventions: National strategies are insufficient; interventions must be subnational (district/provincial).
• Priority Areas: Immediate focus is needed in the Western, North-Western, and Luapula provinces.
• Integrated Approach: Programs should combine malaria control, nutritional supplementation, and improved healthcare access, specifically targeting HIV-positive women, pregnant women, and rural populations.
• Financial Access: Removing financial barriers to medical care is a critical lever for reducing anemia severity.

Limitations:
The study is cross-sectional (limiting causal inference), relies on DHS data which lacks specific biomarkers (e.g., malaria infection, BMI, hookworm), and uses displaced GPS coordinates which may slightly underestimate hotspot intensity.