In Brazil’s Amazon, AI is making healthcare safer

The Amazonian municipality of Caracaraí has 22,000 inhabitants and an overworked pharmacist named Samuel Andrade.

Andrade arrives at work at 8 a.m. to handle hundreds of prescriptions from free government clinics. Most days, he can’t get through all of them. He sometimes gets stuck for hours cross-checking drug databases to ensure nothing has been prescribed incorrectly by rural doctors. 

It is stressful work. He has to help the dozens of patients who line up at his dispensary every day, some of whom have traveled for days to get there. Sometimes he has to rush through prescriptions, and worries he will miss something dangerous.  

In April, Andrade welcomed a new assistant: artificial intelligence software that flags potentially problematic prescriptions and digs up the data to help him decide if they are safe. It has quadrupled his capacity to clear prescriptions, he told Rest of World. In the months since he started using the AI assistant, it has caught more than 50 errors, he said.

“It works instantly and uses digital rather than physical reports,” said the 34-year-old. 

• 

Pharmacists in Brazil began testing the technology earlier this year. Its initial success suggests it could be a game changer for the country’s overburdened primary care system. 

The South American nation has the world’s most ambitious state-backed universal health-care program, by some measures. It offers health care to everyone, everywhere, for free — even to non-citizens. It strives to serve more than 200 million people, which means most of the tens of thousands of government facilities are overstretched, particularly in rural areas where there are fewer doctors and pharmacists. 

The pharmaceutical AI assistant was developed by a Brazilian nonprofit called NoHarm, established with grants from big tech firms including Google and Amazon. 

NoHarm was the brainchild of two smart siblings: Ana Helena, a pharmacist, and her brother Henrique Dias, a computer scientist. They began working on a tech fix for the nightmare of paperwork and research required of pharmacists, often over Sunday lunches at their family home in the southern city of Porto Alegre.

They built an open-source machine learning model trained on thousands of real-world examples. To train the software, NoHarm gathered anonymized patient data and historical prescriptions. The founders fed the algorithm thousands of real-world drug combinations, dosage errors, and adverse interactions. Over time, it learned to spot patterns that even experienced professionals might miss, Dias, also the company’s chief executive officer, told Rest of World. 

It can process hundreds of prescriptions at a time, identifying potential red flags, the company said. NoHarm is designed to support decision-making, not make decisions. 

• 

• 

“We provide a set of alerts, and the professional will evaluate them through the lens of the patient’s needs,” Dias said. 

Over the last seven years, NoHarm has received grants and donations of cloud services, software, and other support from Google, Oracle, Nvidia, Amazon, and the Gates Foundation. It charges private health-care facilities for its software while offering it to public-sector facilities at no cost. 

Brazil has been a latecomer to the AI party. While hospitals around the world have been adopting AI, only 1% of the more than 50,000 public health-care units in Brazil used it last year, according to a report from the Regional Center for Studies on the Development of the Information Society. 

“There are other actors ahead of us,” sociologist Rodrigo Brandão, an author of the Center’s report, told Rest of World. “We’re still learning to walk.” 

Brazil’s unified health system could, however, eventually give it an edge in developing AI for health care, as the country has a uniquely large and diverse database to use for training AI. Unlike countries with fragmented systems or privatized data, Brazil’s centralized records offer a rare opportunity to integrate both public and private data sets and test AI tools at scale.

“That’s very strategic for AI development,” said Brandão. Around 20 cities in the country’s poorest regions have begun using the NoHarm software. It is being adopted in the northeast and deep in the Amazon.

Caracaraí has become an unlikely early adopter. 

While the municipality spans the Amazon with an area larger than the Netherlands, its biggest settlement is a patch of single-story homes, a few stores, and mostly dirt roads. It is inhabited chiefly by fishing and farming families.

Many in the region are underemployed and rely on government relief funds to support their households. Many inhabitants live deep in the Amazon and must travel for days by boat to obtain their prescriptions.

Andrade works out of a small dispensary, filled with shelves and cabinets of various medications, piles of paperwork, and a computer. He hands out prescriptions through a small window to the long line of people who arrive every day. 

The patients often have a combination of chronic conditions that require multiple drugs, making pharmaceutical review complicated.

“Many things slip past our eyes, or we simply don’t know,” said Andrade. “The system lets us cross-check information much faster.”

Between checking and handing out prescriptions and managing inventory, he is overworked and exhausted most days. In his five years since moving to Caracaraí to serve as the main pharmacist for the region, he hasn’t had time to visit the other outposts in the municipality where drugs are distributed.

Until this year, much of his day was spent looking up information on drugs online to cross-check prescriptions. It was challenging, he said, given the unreliable internet and uncertainty about where to find the latest information. 

Today, he just has to log in, and the software lists all the prescriptions he has to handle. It clears the ones that are safe and flags those that need further investigation. 

NoHarm highlights potential problems with labels like “medicine interaction” and “overdose.” It provides links to the data sources on which the warnings are based.

The NoHarm system has only been used in Amazonian towns for a few months now. It sometimes fails to connect to the internet to come up with answers. 

Some experts are also concerned that health professionals aren’t prepared for the rapid adoption of AI. Critics argue there is a lack of training on how these assistants operate, and workers may not know how to identify bugs, hallucinations, or errors in the software. 

“There’s a challenge for not only learning how to use this tech, but also on how to direct IT professionals to improve it, since it demands continuous adjustments and tunings,” said Brandão.

NoHarm says its AI assistant is designed for transparency: Each flagged alert includes a reference to a medical source, and pharmacists are encouraged to override suggestions based on patients’ needs. The cultural shift toward trusting AI to assist in health care decision-making may take time.

The startup said the software is still a work in progress and is continually improving. Its original training was mostly based on prescriptions from hospitals in the south of the country, which has different illnesses than Caracaraí. Certain tropical diseases, such as malaria, which are very common in Amazonian areas, are rare in Porto Alegre, the town where NoHarm is based.

“Our algorithms may have a ‘disease bias’ where it won’t detect adverse events more common in the north than the usual ones from the south,” said Dias. 

The algorithms can be quickly updated as soon as the AI assistant is trained on data from new regions and feedback from local pharmacists, he said. 

Nailon de Moraes, a rural physician, told Rest of World he appreciates the extra support with AI. He handles boatloads of villagers every day in a small clinic near the Branco River, which runs through Caracaraí. Often, they are in desperate need of medication.

The NoHarm software recently flagged four of his prescriptions as potentially dangerous. He had prescribed too high a dose of one medicine and didn’t realize that the two medicines he prescribed to another patient needed to be taken at separate times in the day. 

NoHarm stopped him from hurting his already ill patients, said de Moraes. 

“This job is a lot to handle, especially for those of us who are younger,” and less experienced, he said.