Improving estimation of vaccine effectiveness during outbreaks in low-resource settings: A case study of oral cholera vaccination during the 2022-2023 cholera outbreak in Malawi

This study demonstrates that by analyzing routine surveillance data from the 2022-2023 cholera outbreak in Malawi using an EpiEstim framework adjusted for WASH interventions, researchers successfully estimated that oral cholera vaccination contributed to a substantial reduction in transmission, achieving an adjusted effectiveness of approximately 62%.

The Gist

The Big Picture: Fighting a Fire in a Crowded City

Imagine a massive, crowded city (Blantyre, Malawi) where a dangerous, invisible fire (Cholera) suddenly starts spreading. The fire spreads through dirty water and poor sanitation. To put it out, the city leaders had two main tools:

1. Water and Hygiene (WASH): Cleaning the streets, fixing pipes, and giving people buckets and chlorine to clean their own water.
2. The Vaccine (OCV): Giving people a "force field" (a shot) that makes them immune to the fire.

The problem? The fire was spreading fast, the city didn't have perfect records of who got the shot, and the "firefighters" (health workers) were using both tools at the same time. It was very hard to tell: Did the fire go out because of the shots, or because we cleaned the streets?

This paper is like a detective story where the researchers built a special "mathematical microscope" to figure out exactly how much the vaccine helped, even when the data was messy.

---

The Detective Work: How They Did It

Usually, to prove a vaccine works, you need to interview thousands of sick people and ask, "Did you get a shot?" But in a chaotic emergency like this, that's impossible. Many people didn't get tested, and records were incomplete.

Instead, the researchers used a clever trick involving traffic flow.

• The Analogy: Imagine the cholera outbreak is a traffic jam.
  • $R_t$ (Reproduction Number): This is a number that tells you how many new cars (new cases) one sick person is causing. If $R_t$ is 2, one sick person makes two more sick people. If it drops to 1, the traffic jam is stable. If it drops below 1, the traffic jam clears up.
  • The Strategy: The researchers watched how the "traffic jam" ($R_t$) changed day-by-day. They then looked at a graph showing how many people had received the "force field" (vaccine) over time.

They asked a simple question: "As the number of people with force fields went up, did the traffic jam get smaller?"

They also had to account for the street cleaning (WASH). They built a model that said, "Okay, we cleaned the streets on these days. Let's subtract that effect so we can see what the vaccine did alone."

What They Found

Here are the results, translated from "science-speak" to plain English:

1. The Fire Was Real: The outbreak started small, then exploded in late 2022. It was mostly affecting men in the busy city center.
2. The Vaccine Was a Hero: Even without adjusting for the street cleaning, the researchers estimated that the vaccine reduced the spread of cholera by about 53%.
   • Think of it this way: If 100 people were about to catch cholera from a sick neighbor, the vaccine stopped about 53 of them from getting sick.
3. The Street Cleaning Helped, But... The street cleaning (WASH) did help a little bit, but it wasn't the main reason the fire stopped.
4. The Real Power: When they mathematically removed the effect of the street cleaning to see the vaccine's pure power, the vaccine looked even stronger. It was responsible for reducing the spread by about 62%.

The Takeaway: The vaccine was the heavy lifter. It did the heavy work of stopping the outbreak, even while the street cleaners were doing their part.

Why This Matters (The "So What?")

In the past, if a country didn't have perfect computer records or enough doctors to interview patients, they couldn't prove if a vaccine worked during an emergency. They often had to guess or wait years for data.

This paper is like handing health officials a new, portable toolkit.

• The Old Way: "We need to build a massive lab, interview everyone, and wait two years to know if the vaccine worked." (Too slow for an emergency).
• The New Way (This Paper): "We can use the daily reports we already have (who got sick, how many shots were given) and run a quick math model to tell you right now if the vaccine is working."

The Bottom Line

The researchers showed that in a low-resource setting (where data is messy and resources are tight), you don't need perfect data to make good decisions. By using a smart mathematical approach, they proved that the oral cholera vaccine was highly effective in Malawi.

It's like saying: "Even though we didn't have a perfect scoreboard, our math shows that the vaccine was the MVP (Most Valuable Player) in stopping this outbreak." This gives governments the confidence to use vaccines quickly in future emergencies, saving lives faster.

Technical Summary

1. Problem Statement

Estimating the effectiveness of Oral Cholera Vaccines (OCV) during outbreaks in low- and middle-income countries (LMICs) is methodologically challenging.

• Data Constraints: Vaccination campaigns are often implemented rapidly over short timeframes, while individual-level data (e.g., vaccination status of cases) are frequently incomplete due to weak surveillance infrastructure.
• Design Limitations: Traditional study designs, such as case-control or cohort studies, are often infeasible during emergencies because few vaccinated cases are observed, and transmission dynamics change faster than recruitment periods.
• Confounding Factors: Outbreak responses usually involve concurrent Public Health and Social Measures (PHSMs), specifically Water, Sanitation, and Hygiene (WASH) interventions. Most existing studies fail to adjust for these concurrent measures, potentially biasing vaccine effectiveness (VE) estimates.
• The Gap: There is a critical need for pragmatic, real-time approaches using routinely collected data to generate policy-relevant evidence on vaccine impact in resource-limited settings.

2. Methodology

The study utilized a novel ecological modeling approach linking time-varying transmission dynamics to intervention coverage.

• Study Setting: Blantyre District, Malawi, during the 2022–2023 cholera outbreak (January 2022–April 2023).
• Data Sources:
  • Case Data: Routine surveillance line-lists from 32 health facilities (8,013 total cases).
  • Vaccine Data: Administrative records of OCV doses administered (Euvichol-Plus), calculated as cumulative coverage per eligible population.
  • WASH Data: Daily counts of households reached via Case-Area Targeted Intervention (CATI) strategies (disinfection, water treatment, hygiene kits).
• Analytical Framework:
  1. Time-Varying Reproduction Number ($R_t$): Estimated daily $R_t$ using the EpiEstim package in R, assuming a gamma-distributed serial interval (mean 5 days, SD 8 days).
  2. Breakpoint Identification: Used the strucchange package to identify the transition from sporadic to epidemic transmission (identified as December 11, 2022). Analysis was restricted to the epidemic phase.
  3. Regression Modeling: A log-linear framework was used to model $R_t$ as a function of cumulative vaccine coverage and WASH intensity.
     • Approximation: The model assumed that for low coverage, the log of $R_t$ is linearly related to coverage: $\ln(R_t) \approx \ln(R_0) - \beta \times \text{Coverage}$. Here, the coefficient $\beta$ directly estimates VE.
  4. Lag Structures: Applied a 7-day biological lag to vaccine coverage (time to immunity) and a 7-day rolling average for WASH (reflecting gradual environmental improvement).
  5. Uncertainty Propagation: Used posterior sampling (1,000 draws) from the $R_t$ distribution to propagate uncertainty into VE estimates.
  6. Sensitivity Analysis: Tested alternative lag structures (0 and 14 days) and cumulative WASH exposure to assess robustness.

3. Key Contributions

• Methodological Innovation: Developed a pragmatic framework to estimate population-level VE during active outbreaks where individual-level designs are impossible. It leverages the relationship between $R_t$ and coverage rather than individual case histories.
• Adjustment for Confounders: Unlike previous studies, this approach explicitly adjusts for concurrent WASH interventions, isolating the vaccine-attributable effect from environmental improvements.
• Real-Time Applicability: Demonstrated that routine surveillance data (case counts, administrative coverage, intervention logs) can be used to generate timely, actionable evidence for Ministries of Health without requiring new data collection.
• Integration of Evidence: Combined spatial, genomic, and transmission dynamics data to provide a holistic view of the outbreak control in Malawi.

4. Key Results

• Outbreak Dynamics: The outbreak peaked in December 2022. Following the peak, transmission declined steadily through April 2023, coinciding with the rollout of reactive OCV and WASH interventions.
• Demographics: Cases were concentrated in urban Blantyre, predominantly affecting middle-aged men (median age 23 years).
• Vaccine Effectiveness (VE):
  • Unadjusted VE: Estimated at 53.5% (95% CrI: 42.5–64.1%).
  • WASH-Adjusted VE: After adjusting for the 7-day rolling average of WASH activity, VE increased to 62.1% (95% CrI: 49.3–74.9%).
  • Interpretation: The increase in VE after adjustment suggests that WASH interventions had a modest independent effect, and the decline in transmission was not solely driven by WASH; the vaccine contributed significantly and independently.
• WASH Impact: The WASH intervention alone showed a small reduction in transmission intensity (Hazard Ratio median: 0.993), though this was statistically weak, likely due to intermittent deployment and the difficulty of measuring environmental uptake via administrative logs.
• Sensitivity: Estimates remained robust across different lag structures, though a 14-day lag yielded lower VE (35.9%), highlighting the importance of appropriate biological lag selection.

5. Significance and Implications

• Policy Relevance: The study provides a validated method for health authorities in LMICs to evaluate intervention impacts in real-time during emergencies. This allows for rapid decision-making regarding the need for additional vaccine rounds or intensified WASH efforts.
• Validation of OCV: The findings confirm that reactive OCV campaigns are highly effective in reducing cholera transmission even in complex, resource-constrained settings with concurrent interventions.
• Future Research: The paper highlights the limitations of relying solely on administrative WASH data and calls for better integration of behavioral and environmental data in future outbreak evaluations.
• Global Health: This approach offers a scalable solution for estimating vaccine impact in other infectious disease outbreaks where classical epidemiological designs fail, bridging the gap between emergency response and evidence generation.

Conclusion: The study successfully demonstrates that by modeling the time-varying reproduction number against intervention coverage, it is possible to generate robust, adjusted estimates of vaccine effectiveness during active cholera outbreaks, providing crucial evidence for optimizing public health responses in low-resource settings.