1. A smarter way to count carbs

For people with diabetes, meals can be a challenge. Patients must figure out how many carbohydrates each food contains and then counterbalance it with the correct dose of insulin. If they don’t, their blood glucose levels fluctuate dangerously.

For the past twenty years, continuous glucose monitors have helped. These devices allow insulin to be administered closer to the time it’s most needed. But patients still must manually count carbs and calculate the dose for every bite they eat.

Now, a new AI system from Stevens Institute of Technology can detect when a person is eating and calculate how many carbohydrates they’re taking in. The algorithm mines a library of graphs that represent glucose level changes for each food consumed and then delivers the correct amount of insulin.

This smart system models carbohydrate intake accuracy to within 1.2 grams. Previous systems have an error rate of 17 grams or more. And because this AI can make estimates very quickly, the insulin delivery times are cut nearly in half.

The researchers now want to access larger data sets to create more accurate simulations. They also hope future versions will incorporate data from other circumstances that affect glucose level, such as exercise, hormones, sleep, and stress.

2. Viewing disease from above

Schistosomiasis is a parasitic disease that affects 240 million people worldwide, second only to malaria in its global impact. Caused by a flatworm that lives inside the human circulatory system, it causes abdominal distention and ultimately bladder or liver cancer.

Though treatable with drugs, people in at-risk communities are easily re-infected when they come into contact with larval forms of the parasite in the freshwater where they bathe. Diagnosing people individually is time-consuming and labor-intensive, especially in remote villages.

That’s why Chelsea Wood, a parasite ecologist at University of Washington, wants to find an easier way to identify transmission hotspots. The schistosomiasis parasite requires a type of snail to complete its life cycle, and more snails in the water means a higher risk of infection.

Wood started her study by counting and mapping snail distribution at 30 sites in northwestern Senegal. But snail populations were distributed unevenly and inconsistently over time. It turns out that the snails preferred a specific habitat of floating vegetation—one that stands out as dark green or brown on aerial drone images.

Wood’s group then developed models to predict schistosomiasis transmission, reducing to a fraction the time needed to evaluate risk. The team also hopes to add a machine learning component to automate the identification of snail habitats in aerial or satellite photos, leaving more resources available for treatment. 

3. Getting a clue about women’s bodies

Medical research has a problem: most of it is based on men’s bodies. This means that women are less likely to be diagnosed with serious conditions, preventative methods are less likely to work, and symptoms are ignored.

So it’s probably no surprise that when it comes to menstruation and pregnancy, the information gap is even wider. Women don’t know if what they experience is normal. And, sadly, neither do medical professionals.

But that may change, thanks to the rise of personal health apps that are giving researchers unprecedented access to large datasets about women’s bodies.

Several research studies are now underway using data collected from Clue, a period-tracking app. For example, computational biologists at Stanford are drawing on the data to research menstrual pain patterns. They hope this will help to identify disease risk groups and provide early warnings for underlying medical conditions.

Elsewhere, scientists in the School of Public Health at Columbia University are combining traditional research methods with app-sourced period data to explore associations with breast cancer. Information volunteered by cycle-tracking apps lets researchers collect real-time daily information they hope will provide a clearer picture than previous studies hampered by biased and out-of-date methods.

Another Columbia University study uses machine learning to discover new insights about health and behavior. By identifying patterns in the tracked data, they are hoping to identify phenotypic groups who experience similar symptoms. This same group is also using a separate app to study endometriosis – a painful and incurable condition that afflicts 1 in 10 women.

4. Paging doctor data

Rushing from one emergency to another, sleep-deprived and overworked – that’s a typical shift for a medical resident. Once seen as a rite-of-passage, the medical community has recently acknowledged that the long hours residents put in can lead to serious medical mistakes.

But Amit Kaushal of Stanford University believes that problems arise not from how long a resident is on shift, but what happens while they’re there. To test that theory, he collected two years of pager data from Stanford residents, more than half a million pages in total.

The results showed that a slow night might mean only 10 pages, but a busy one could result in more than 80. The pager results were checked against patient admissions data—longer wait times are associated with high hospital workloads—to confirm that more pages really did mean busier residents. 

Kaushal hopes that investigating this digital data stream will lead to more accurate evaluations of resident workloads and a future in which, instead of just reporting retrospectively, they can be managed and adjusted in real time.

We can probably all agree that we would prefer that the resident who treats us in the hospital be as alert and well-rested as possible.