Artificial intelligence inserts our lives in many ways. While new technologies always have their advantages and disadvantages, the disadvantages of using AI can be disastrous for individuals prone to mentally ill episodes, they can also be used to predict when and how mentally ill people will respond to interventions designed to be useful.
Large-scale language models (LLMs) like ChatGPT are a type of machine learning that uses human languages trained in many other language datasets to provide results and answers in the same format. LLMs can take the human voice, so they often are human-like or look like humans.
The type of machine learning used in scientific research differs from LLM. Machine learning, used in mathematical or scientific fields, takes data (such as demographic data, measurements, test scores) and provides predictions based on the desired goals that the user will prompt and display as numerical information rather than language. Machine learning used in this way can be a useful tool for predicting what can be done to help people. These predictive models typically focus on preventive and positive goals, such as preventing recurrence. It will probably predict who will suffer mental illness in the first place.
Now we are in the process of revealing what these models can communicate to us. So far, one thing that can be done is predicting how patients with schizophrenia, a very serious mental illness, will respond to treatment like ongoing drug regimens. That was exactly what a group of researchers did in a study published earlier this year.
How the research was set up
Korean researchers have created the algorithm with the in mind predicting who will respond to treatment plans. Their idea was to identify the type of patient who responded best to a variety of options. By designing new predictive models, machine learning methods can understand people with individualistic traits, those who can advance the use of psychodiagnosis. By personalizing patients, we can move from a wide range of categories to specific concerns in the hopes of improving treatment outcomes.
To that end, researchers obtained data from 299 people. Almost half of them were women, with an average age of 36.7 years old. The average time the sample was already under treatment was 8.8 years. 33% of the samples also suffered from the metabolic syndrome. This is a health problem that occurs more frequently in schizophrenia patients.
Over 24 weeks, patients were instructed to switch from a single drug to an expanded release drug. The drugs used were well-known antipsychotics such as risperidone and paliperidone. Usually, patients can take one pill that paces the release of chemicals in the day, making it easier to take an expanded release medication (as opposed to taking it twice a day for more immediate effects).
Starting new schizophrenia medications or moving from one species to another requires real commitment as it is not easy. In this setting, those who are reluctant to continue taking the medication may drop out of the study and return to their old treatment plan. If this occurs in a real clinical setting, the patient may ultimately stop the medication.
So, what factors can help determine if a patient will drop out? To identify factors that influence whether patients continue treatment, the researchers followed several factors, including sociodemography, psychosocial functioning, attitudes towards medication, and metabolic and endocrine health characteristics as predictive properties.
What researchers found about treatment retention
It was found that BMI was the best predictor of treatment response, followed by attitudes towards medication. Positive attitudes regarding drug treatment were associated with ongoing commitment. This is a connection that can be easily guessed by reading countless anti-psychiatric stories online. This was true at each measurement and follow-up of the study at a 4, 8, and 24-week assessment of patient responses to drug continuation.
The researchers did not identify any causal explanations that could explain why BMI is a predictor of ongoing treatment outcomes. However, it may suggest that attention to health and fitness baselines may contribute to how we approach drug therapy.
Factors predicting treatment response early in the study included personal and social relationships and the severity of negative symptoms. The 24-week mark, education level, severity of positive and negative symptoms, and duration of illness were associated as important predictors of treatment response.
Where machine learning can be entered
The researchers' goal was to create more cost-friendly practical factors (compared to neuroimaging and genetic data) to determine whether patients should continue treatment. For example, factors such as BMI and medication attitudes are easy to assess within clinical office settings.
If these can predict treatment outcomes, patients and physicians can save time by starting with the type of treatment determined by machine learning tools to be the most effective for patients with a specific demographic. Treatment is highly personalized and the field of psychiatry recognizes it all around the world. Machine learning can be used as a tool to assist patients rather than sending them to episodes.
Currently, machine learning technology is only available to researchers. However, the goal of these early research is to make its application relevant to daily life. Machine learning, and perhaps the general artificial intelligence, can do more to patients when using it for more productive reasons, such as recovery goals. With LLMS, when a user searches for the best treatment for psychosis and prompts, it may produce useful results. LLM does not generate numerical data in an academic sense, but can point out useful information about existing data regarding recovery. The idea is to use technology to help people. Avoiding falling into the trap of enabling AI will help us guide our goals and outcomes.
