Counting to 73,700: A Guide to Randomization

Google Earth map of Shakawe, Botswana

From: Scott Dryden-Peterson
Sent: Friday, June 19, 2015
Subject: BCPP milestone

Today we completed mapping of the last study community. In one of the many remarkable behind-the-scenes contributions that has made a project of this scale possible, during nights and weekends over the past 18 months, Oaitse (cc’d here) single-handedly identified and labeled ~73,700 households from Ranaka to Shakawe. We are indebted to you, Oaitse.


Why Randomize?

Randomization is a key element in the best clinical trials. It insures that researchers don’t inadvertently introduce their own preferences or biases into the hard work of trials. It helps safeguard the integrity of the data.

In the Botswana Combination Prevention Project (BCPP), randomization occurs at several levels. In each of the 15 pairs of villages in the trial, one village gets the combination prevention interventions and the other village acts as a control. In each pair, a lottery system was used to decide which village received the combination interventions. Village names were placed inside plastic, egg-like containers and randomly selected in the presence of Ministry of Health officials and independent observers. But randomization doesn’t end there.

Researchers hope the BCPP will show how to reduce new HIV infections at a community level. Testing for incidence (the number of new infections in a year) has always been a challenge in AIDS research. To do it accurately involves labor-intensive testing of a large number of people year after year to determine how many are newly infected.

The total population of all 30 BCPP villages is about 180,800. Of those people, about 105,000 are age-eligible (16–64) for the trial. Rather than testing all 105,000 people, the BCPP team is testing 20% of them. But for that 20% to act as a proxy for the entire population, the people must be randomly selected. That crucial process involves a number of steps.

The Best Woman for the Job

One of the first steps was using Google Earth to number the households in all 30 villages. Households in Botswana usually include a compound with a number of dwellings, either rondavels (round traditional thatch-roofed huts) or houses built with cement blocks. Some compounds include covered enclosures for animals. It takes a discerning human eye to look at Google Earth satellite images and decide where one household ends and the next begins. And that’s just how Oaitse John spent her evenings and weekends for several months.

Oaitse John. Photo by Dominic Chavez

Oaitse, who worked fulltime at the Botswana Harvard Partnership (BHP), was approached by Dr. Scott Dryden-Peterson, a Harvard researcher and BCPP investigator. “Oaitse is detail-oriented and accurate,” said Scott. “But she also runs a hot dog stand and a couple of small businesses. She’s an entrepreneur.” When Oaitse heard about the possibility of working after hours on the BCPP, her reaction was, “Yes, the big project. Bring it on.”

Though Oaitse didn’t have a good sense of geography and had never heard of Google Earth, Scott showed her how to open up a village map, zoom in so that individual buildings were visible, and use keyboard commands to insert a numbered flag at the front door of the main house in each compound. The flag automatically assigned latitude and longitude coordinates to the front door. On a trial run, Oaitse identified over 100 households in 30 minutes. “That’s how she got the job,” said Scott.

Defining Households

Though the instructions were simple, defining households from a satellite map was a complicated process that involved knowledge of village culture, a fair amount of good judgment, and some guessing.

In a traditional Botswana household, several buildings are often enclosed by a fence to define the compound. The fence helps with security, reinforces land boundaries, and keeps animals from trampling plants in yard. But not all households have fences and sometimes two or more households are contained within a fenced area.

Shakawe household
Shakawe household. Photo by Molly Pretorius-Holme

Multigenerational living is also the norm in Botswana villages. Oaitse, for example, lives in her grandmother’s house in Ramotswa, a town near the South African border. Her uncles, sisters, and nephew live in the same compound, but different houses. If new houses are built, old houses are often left standing.

All these factors made the task of numbering households, or “plots” as they are called in the study, anything but simple.

Working on a MacBook Air, Oaitse numbered plots during her lunch breaks. At the end of a workday, she’d often stay in the office and number for a few hours, go home, take a shower, get something to eat, and number more plots before bedtime.

She mapped on the weekends. Her sisters complained that she was always working. “It was a little bit addictive,” Oaitse admitted, like the video games she played on her phone. She got frustrated when the Internet was slow or not working at all. When Scott opened Google Earth to review her work, he could see the late hours she was keeping.

Oaitse had to keep ahead of the BCPP rollout. As the field team completed their work in one pair of villages, plots in the next pair had to be randomly selected before the team could begin work. The random selection depended on Oaitse’s numbering. She kept pace.

After she finished numbering all the plots in a village, the file was sent to Scott for review, then on to the IT department at the BHP, where the data was entered into the EDC (Electronic Data Collection) system, which renumbered each household with a unique study ID. The file was then forwarded to a senior epidemiologist who ran an automated script to randomly select 20% of the numbered households for the field team to visit, plus another 5% to replace any households in the 20% that were uninhabited or whose members chose not to participate.

A Research Assistant locating households.
A Research Assistant locates households with a MacBook AIr. Photo by Dominic Chavez

When the field team arrived in a village to conduct the Baseline Household Survey, the coordinates for each plot were loaded onto GPS receivers. With GPS in hand, Research Assistants were assigned to find specific households. If things went according to plan, at about five meters away from the front door—the point where Oaitse placed her virtual flag—the GPS screen flashed the message, “Arriving at destination.”

Needless to say, things didn’t always go as planned. When you’re numbering 73,700 plots, there are bound to be discrepancies between the map on a computer screen and reality on the ground. Some households turned out to be churches or businesses. Some plots were deserted.  But the system, created from scratch, worked surprisingly well.

“It was exciting to do the mapping,” said Oaitse. She sometimes runs into members of the field team at BHP headquarters in Gaborone. “Some of the research assistants would talk about which village they were working in and I’d be like, ‘I know that village. I’m the one who mapped it.’”

The mapping also gave her a wider sense of the world. “I felt like it would be great for me to go and physically see these places, to go around and see my country and how people are living. I need to create some time and save some money and travel around and see.”

Title Image: Households in Shakawe mapped with Google Earth