It’s no secret that the U.S. spends a lot on health care, around 18 percent of its GDP or $9,400 per capita, nearly double what other high-income countries such as Canada, UK, Germany, and Australia spend.

But more spending doesn’t necessarily yield better results. In fact, studies show that many of the countries that spend less than the U.S. see better outcomes in the overall health of their citizens. According to a new report published by the Journal of the American Medical Association (JAMA), a little less than half the health care expenditures in the U.S. go into planning, regulating, and managing medical services at the administrative level.

And industry experts believe we can reduce a lot of this spending with the help of artificial intelligence. “A number of recent studies have found that [health care] administrative costs [in the U.S.] continue to rise and or remain higher than other countries,” says Pamela Hepp, an expert in data security, health care regulation, and digital health records at Buchanan, Ingersoll & Rooney. “There is always room for improving efficiencies in the delivery of care and AI has some promise in that regard.”

Artificial intelligence has already made great progress in the field of medicine, helping in tasks such as processing x-ray images and detecting cancer and assisting doctors in diagnosing and treating patients. “I am not convinced that we will see AI replacing physicians, physician extenders, or other health care professionals any time soon,” Hepp says, referring to the general fears surrounding automation. “However, AI may be useful in helping to reduce administrative costs and lead to economies of scale.”

Population health management

Among the domains where Hepp believes AI can help reduce costs is population health management, the discipline within the healthcare industry that studies and facilitates care delivery across the general population or a group of individuals. Regulations such as Health Insurance Portability and Protection Act (HIPPA) and the HITECH Act have helped propel the digitization of the health care industry and incentivize the development and adoption of electronic health records (HER). However, the real value that lies in the reams of medical data organizations and companies collect across the population remains elusive.

“Population health is not a destination. It’s a moving target whose goals, once achieved, will be quickly replaced with new ones, just out of reach,” writes Prashanth Kini, the vice president of product management for health care at machine intelligence software company Ayasdi.

The complex and dynamic nature of population health management makes it especially convenient to handle with machine learning, the branch of AI that specializes in uncovering hidden patterns and predictive trends in large and disparate sources of data in short timespans.

Current population health management tools rely on analysts querying health care datasets. However, “there are countless patterns and trends that won’t be uncovered because clinicians failed to ask the right question,” Kini says.

AI can bridge this gap through “unsupervised learning,” a subset of machine learning that analyzes data and discovers common patterns and anomalies with minimal human involvement. AI algorithms powered by unsupervised learning can ingest data from health records, financial data, patient-generated data, IoT devices, and other relevant sources to automatically discover groups of patients that share unique combinations of characteristics.

The patterns gleaned from population health data can help health institutions engage in preventive and predictive care, which can result in great savings in managing and treating diseases that become costly and complex when they’re discovered at later stages.

Evidence-based medicine

“[AI] may be used in the identification or development of evidence-based medicine and treatment protocols that can be used, generally, for the treatment of specific diseases,” Hepp says.

Evidence-based medicine is the practice of making treatment decisions based on observations that come from clinical studies of populations. While evidence-based medicine is not new, the speed of human processing power previously limited its application. Today, AI algorithms are helping boost those efforts by analyzing millions of data points and quickly finding relevant patterns and courses of action.

This approach can make a big difference both in costs and patient health when diagnosing and treating diseases such as cancer. “Using consensus algorithms from experts in the field, along with the data that oncologists enter into a medical record (i.e., a patient’s age, genetics, cancer staging, and associated medical problems), a computer can review dozens, sometimes hundreds, of established treatment alternatives and recommend the most appropriate combination of chemotherapy drugs for a patient,” writes Stanford University Professor Dr. Robert Pearl in Forbes.

As an example, Pearl names the work done by researchers at the Permanente Medical Group’s Division of Research, who used AI algorithms and data collected from 650,000 hospitalized patients to create a predictive model that could identify which hospitalized patients today are most likely to end up in the ICU tomorrow. This approach not only helps reduce the costs of repeat hospitalizations and releasing at-risk patients early, but it also saves numerous lives each year.

Medication research and discovery

Another area that Hepp believes AI can help reduce costs is the development of new drugs and vaccines. Discovering new drugs is time-consuming and expensive and involve thousands of researchers who conduct experiments to find the viable solution that can slow, stop or reverse the effects of a disease. The process can take up to 12 years and cost more than a billion dollars.

AI can help slash expenditures and speed the drug development process by boosting analysis and research efforts. An example is a partnership between Pfizer and IBM Watson to apply AI in the drug research process. IBM will use the massive computational power and cognitive abilities of its AI platforms to quickly analyze and test hypotheses from “massive volumes of disparate data sources” that include more than 30 million sources of laboratory and data reports as well as medical literature.

For instance, researchers can use Watson’s natural language processing (NLP) capabilities to analyze the content of thousands of medical papers and reports at very fast speeds to find facts that are relevant to their work.

Hurdles remain

While AI shows great promise in optimizing the administration of health care services and work at different levels, it must still overcome challenges. “The deployment AI in and of itself may be costly,” Hepp says. “Technology is one of the factors that continues to impact administrative costs. Moreover, its use will not be without regulatory hurdles that may slow the adoption and impact overall cost savings.”

For instance, the introduction of any new technology may require FDA approval. Moreover, the collection and processing of patient data must be compliant with HIPPA and other digital privacy-related regulations.

AI will also have to overcome cultural hurdles. “Unfortunately, the biggest barrier to artificial intelligence in medicine isn’t mathematics,” Dr. Pearl says. “Rather, it’s a medical culture that values doctor intuition over evidence-based solutions. Physicians cling to their independence and hate being told what to do. Getting them comfortable with the idea of a machine looking over their shoulder as they practice will prove very difficult in years to come.”

Ben Dickson is a software engineer and the founder of TechTalks, a blog that explores the ways technology is solving and creating problems.