Spatial and Machine Learning Analysis of Breast and Cervical Cancer Screening Uptake in Ghana: Evidence from the 2022 Ghana Demographic and Health Survey

In many parts of the world, finding cancer early is the difference between life and death. For women in Ghana, two types of cancer—breast and cervical—are major health concerns. The most effective way to catch them early is through screening: a simple check-up where a doctor examines the breast or tests the cervix. However, getting these services is not the same for everyone. Past studies have shown that wealth and education act as gatekeepers; women with more money and more schooling are far more likely to get screened than those with less. This pattern is well known, but it has usually been viewed through a broad lens, looking at entire regions or the country as a whole. This approach treats a region like a single block of color, hiding the fact that conditions can change drastically just a few miles away. It also relies on standard statistical methods that assume relationships are simple and straight lines, potentially missing complex ways that different factors interact.

A team of researchers in Ghana decided to look closer. They asked whether the pattern of who gets screened changes in specific, small pockets of land that standard regional maps miss. They also wanted to see if modern computer learning tools could predict who is likely to be screened better than traditional methods, without needing to know the exact location of every woman. By combining detailed maps with advanced computer models, they aimed to create a sharper picture of health inequality, one that could help health officials target their efforts more precisely.

The researchers turned to a massive dataset called the 2022 Ghana Demographic and Health Survey. This survey interviewed thousands of women across the country, asking about their health, their homes, and whether they had ever received a breast exam or a cervical test. The team focused on women between the ages of 25 and 49, a group most at risk for these cancers. They linked the survey answers to the specific geographic locations where the interviews took place. While the exact coordinates were slightly shifted to protect the privacy of the participants, they were precise enough to show patterns at the level of small communities, known as clusters, rather than just large administrative regions.

First, the team mapped the data to see if screening happened in random spots or if it clumped together. They found that it definitely clumped. Areas where women frequently got screened were often surrounded by other areas where screening was also common. Conversely, areas with low screening rates tended to be neighbors with other low-rate areas. This clustering was significant for both breast and cervical cancer, but it was stronger for breast exams. When they zoomed in, they saw "hotspots" of high screening activity concentrated around the capital city, Accra, and parts of the central and southern regions. In contrast, "coldspots" where screening was rare were heavily concentrated in the northern parts of the country. This confirmed that the divide between the north and south is real, but it also revealed that the problem is not uniform even within those large areas; there are specific pockets of need that a broad regional average would hide.

Next, the researchers built a computer model to predict which women were likely to get screened based on their personal circumstances. They fed the model information about age, education, wealth, where they lived, and whether they had health insurance. Crucially, they did not tell the computer the specific location of the women. Instead, they let the model learn the patterns from the personal details alone. To make sure the model was truly learning and not just memorizing the data, they used a clever testing method. They split the data into groups based on the communities, not the individuals. This meant that when the computer was tested, it had to predict the screening habits of women from communities it had never seen before. This prevented the model from using local knowledge it shouldn't have had.

The computer model performed reasonably well, correctly distinguishing between women who were likely to be screened and those who were not with an AUC of 0.71 to 0.73. The most important factors driving these predictions were education and wealth. Women with higher education and more money were consistently predicted to have higher screening rates. Age also played a role, with older women in the group being slightly more likely to be screened. Interestingly, the specific region a woman lived in mattered less once her education and wealth were taken into account. This suggests that the reason some regions look different on a map is largely because the people living there have different levels of education and income, rather than because the region itself has some unique, unexplained quality.

The study also looked at how unequal the access to screening really is. When they calculated the gap between the rich and the poor, they found a surprising twist depending on how they measured it. For cervical cancer, the gap looked huge because very few women overall were getting tested, making the rich appear to have a massive advantage. However, when the researchers adjusted the math to account for the fact that the overall numbers were so low, the picture changed. The adjusted numbers suggested that the gap for breast exams was actually more concentrated among the wealthy than the gap for cervical tests. In other words, while both services are skewed toward the rich, the inequality in breast screening is more tightly packed around the wealthiest women.

Despite these detailed findings, the researchers are careful not to claim they have solved the problem or that their maps are ready to be used as a final guide for health officials. The computer model was good at predicting based on personal details, but it was not tested on completely new areas outside the survey data, so it might not work perfectly if applied elsewhere. The maps showing the hotspots and coldspots were also treated as exploratory; they point to likely areas of need but were not statistically confirmed with the highest level of certainty because the researchers did not apply a strict correction for testing so many small areas at once. Furthermore, the study could not prove that living in a poor area causes a woman to skip screening; it only showed that the two things happen together.

The ultimate value of this work lies in its ability to show the hidden layers of a public health issue. By moving beyond broad regional averages and using tools that can handle complex patterns, the researchers showed that the story of cancer screening in Ghana is more nuanced than a simple north-south divide. They demonstrated that while wealth and education are the primary drivers of who gets screened, these factors are not evenly distributed across the landscape. There are specific communities, even within wealthier regions, that are being left behind, and specific pockets in poorer regions that are doing better than expected. This level of detail offers a new way to think about targeting resources, suggesting that health planners might need to look at the specific makeup of small communities rather than just the average of a large region. The study serves as a proof of concept that combining spatial mapping with modern predictive tools can reveal the fine print of health inequality, offering a clearer, if still imperfect, view of where help is needed most.