The context​

The challenges of access to health care in Malawi are multifaceted and can be attributed to several factors, including a lack of infrastructure, poor road networks and a shortage of trained health workers. Rural and remote areas are particularly affected by these issues: many communities are forced to travel long distances to access basic medical services. This may lead to irregular or unattended health check-ups, resulting in delayed diagnoses, increased morbidity and high mortality rates.

Addressing these concerns, the Malawian Ministry of Health announced in 2019 their intention to build 900 new health clinics across the country so that 95% of the population will live within 5 to 6 km of a health facility. To realise this goal, the MoH, in partnership with private companies, used mobile phone data to determine the optimal locations for new health posts.

How it was implemented

In order to prepare a recommendation for the locations of these new health clinics, the project team used a three-step approach.

Step 1: Data Collection

For this analysis, the project team integrated three data sources:

1. Anonymised mobile phone data (MNO), which included Call Data Records (CDR) and geo-tagged locations of each cell phone tower. To use the private MNO data, Ministry of Health staff needed to ensure the MNO technical staff were trained to ensure the data was properly anonymised. This data then had to be cleaned: inaccurate records were deleted and as this work was done the project team identified quality issues and missing data, which led to recurring work with MNO staff to properly format the exhaustive datasets. Previous research has demonstrated a strong correlation between the density of unique users within a cell tower's catchment area and the corresponding population density, and as a consequence this data provided valuable insights.‍
2. Population density data from WorldPop. Researchers have developed cutting-edge algorithms that are trained on historical census data to project annual population density at a 100-meter resolution. This information was leveraged by the project team to gain insights into the trends and patterns of population density across Malawi.‍
3. Location of existing health facilities, which was provided by the Ministry of Health and UNICEF.

Step 2: Gap Analysis

Using data from WorldPop and the location of existing health facilities and their service areas, the project team was able to create maps which enabled the calculation of the population in each district that did not live within 5 km of a health facility.

Step 3: Population Movements

After identifying where the gaps in health care service were, the team integrated MNO data to consider population growth and migration patterns. More specifically, the anonymised MNO data was mapped to administrative units to calculate the density of individual callers and then match it to population density using WorldPop data. Fluctuations in caller density could be extrapolated to infer population migration tendencies.

The MNO data showed that the population shifts during work times during the week, shifts to the coast and to markets on the weekends, and that migration takes place during the rainy season. Understanding this pattern of short-term population shifts helped the project team predict population densities over time.

How can better data contribute to better policy? ​

Using this data the analysts developed a model that would optimise the allocation of care facilities and clinics. With the added value of the MNO data, the model endeavoured to maximise coverage of currently unserved populations, taking into account population growth and migration patterns.

With this model, the team was able calculate how many people each new health post would serve. Moreover, this model considered the impact of flooding during the rainy season, determining that some of the health posts would be over-burdened as a result, and adjusted the proposed locations based on this analysis. An added benefit is that periodic or close to real-time MNO data could be used in the future to update the model as health posts are constructed.

The data-driven model proposed optimal locations for the 900 new health posts that would enable 95% of Malawians to live within walking distance of a health facility. These findings were compared with Malawi’s Capital Investment Plan, which included a plan for the locations of the health posts and were found to offer greater coverage while reducing the likelihood of individual health posts becoming overburdened.

Implementing the findings

To ensure that these important findings could be applied effectively by policymakers in Malawi, the project team developed an interactive Power BI dashboard, which can provide:

• Estimated population density
• Health post coverage
• Cell phone usage patterns
• Long-term population movements
• Short-term population movements

From the beginning, the Malawian Ministry of Health (MoH) was closely involved in this use case, providing input and directly engaging on the technical and policy levels. To make the project sustainable and replicable, the project team collaborated with government partners in Malawi to incorporate their analysis into the country's systems and provide a list of priority locations for the new health posts. After the analysis was completed, the project team engaged in a series of meetings with the MoH to provide a detailed walk-through of the results.

Where do we go from here?​

In June 2022 the Minister of Health Khumbize Kandodo Chiponda confirmed government financial support for the construction of the first 55 new health posts. In October 2022, Malawi’s President Lazarus McCarthy Chakwera reaffirmed the government’s plans to construct 900 new health posts in total. The project team’s analysis can continue to inform optimal locations for these new posts, but in order for this to be effective in the long term, a “data pipeline” for anonymising close-to real time MNO data will need to be established.

DIAL remains committed to continuing its work with government partners in Malawi and other countries around the world to demonstrate how data for development models like this one can be replicated across development sectors and different regions.