Finding willing and qualifying patients to take part in market research is a known challenge in our industry. There are a few suppliers that everyone knows as “the patient recruiters.” This poses the question of, “Are AI synthetic patients the answer we have been looking for to fill the gaps?”. There are differing options, however two main themes come up often:
- Synthetic patients can help to fill the knowledge gap for data sets that fall short of desired outcomes. For example, a quantitative survey with HR+/HER2- metastatic breast cancer patients to understand their treatment preferences for targeted therapies. Your sponsor wants to capture 5% of the population of ~170,000 women living with mBC in the USA. As most researchers know, getting 8,500 metastatic cancer patients to be able to complete a 15–20-minute survey can be extremely challenging if not impossible. You run the recruitment and get a total of 3,500 completes from multiple agency efforts, how do you conduct your analysis with such a large data gap?
- Synthetic patients will take the personable and empathetic approach to patient recruitment away, as well as cause skewed data. Using the same data above, if you are showing to the 3,500 patients that their responses are being used to build an AI patient model, it takes away the value of what they are providing. Many of the patients that take part do not do it for the honorarium but rather try and help improve the experience and treatments for future patients that are on the same journey. The idea that their responses are to build a machine that will help guide the future of their condition is concerning to them. How do we overcome this?
Let’s begin by talking about the first thought of using synthetic patients to fill gaps in datasets. While I am always a strong advocate of primary data collection with actual humans, I will agree that there are certain instances where this just isn’t possible. The next thought that comes to mind is which data are we training any type of synthetic patient on? If we are strictly using secondary data, the concern is, are we using outdated, historical data? Is the initial data 100% human intelligence and not already been through an AI program? So, you may be asking, what do you suggest that we use?
- In creating a synthetic patient to fill data gaps it is important to still capture the emotional emphasis of actual patients as well as ensuring that the data you are presenting to the study sponsor is current and correct. How do we do this? My thoughts are that secondary data such as claims data, past primary research data, and even social media skimming are important, but that including a part of actual human insights is pertinent to ensuring your data is correct. Human data is to be used as a way of either finding that there are outliers in your data, or drastic changes in data trends. If you find that the data changes are significant, I will not recommend the use of synthetic patients. Alternatively, if you notice a pattern of patient behavior and thinking, it may be a suitable candidate to continue with building an AI model to fill future data gaps.
- The next challenge is losing that personable and empathetic element of patient insights. How do we overcome this? In our earlier scenario, we were concerned about the quality of insights by showing the use of the AI model. The easiest way to overcome this hurtle is transparency and clear communication on the specific use of the model and the value that it will add not just to the study sponsor or researcher, but the patient themselves. I speak to patients almost daily and they understand the knowledge gap that exists and are more than happy to help in any way they can to help fill those gaps. By helping them understand that this is not an attempt to replace them, but rather to amplify their voice, they will be much more likely to give their buy-in.
In summary, as the use of AI and synthetic data grows across the industry we must adapt and evolve how we use these tools and remember the human patients behind the insights being used. The added value of combining AI Synthetic Data with the Human Data from these patients is unparalleled. Looking to the future the uses of AI + human models could expand to help educate new doctors, give patients access to AI navigators to help them understand their diagnosis and available treatments. The uses are endless and will continue to evolve well into the future.
