By Deborah Borfitz

May 13, 2025 | Given information on two blood biomarkers, together with drug type and dose, artificial intelligence (AI) was used to create personalized “digital twins” of cancer patients in an interventional clinical trial at a pair of Singapore hospitals. The goal was to “optimize only their own doses, dynamically,” according to the platform’s led developer Dean Ho, Ph.D., director of the Institute for Digital Medicine (WisDM) at the Yong Loo Lin School of Medicine of National University of Singapore (NUS) as well as head of its department of biomedical engineering.  

The technology, known as CURATE.AI, considered changes in circulating levels of the carcinoembryonic antigen (CEA) and cancer antigen 125 (CA125). The feasibility study enrolled 10 patients with advanced solid tumors where the treatment intent was symptom relief and improved quality of life, and it demonstrated the platform’s adaptability to their clinical situation (npj Precision Oncology, DOI: 10.1038/s41698-025-00835-7). 

The study ran over a two-year period and involved three designated clinician-investigators who generally opted to follow the recommendations of CURATE.AI, which were delivered to them via a clinical research coordinator. The high rate of user adherence might be partially explained by the fact that the physicians were engaged in selecting data and boundaries for CURATE.AI operations, the researchers report. 

In terms of initial dosing decisions, the platform was found to have limited relevance since investigators weren’t using pre-existing population data to derive the dose recommendations for individual patients, says Ho. “The initial doses given to each patient represent a [structured] calibration process... as part of the CURATE.AI workflow” for creation of patients’ digital twins, which in the future could possibly be embedded into the electronic health record. 

“For some patients, the CURATE.AI-recommended doses were lower [on average] than the standard-of-care guidelines, and/or conventional maximum tolerated dosing,” Ho says. “This early finding suggests that this platform may be effective at pinpointing the optimal dose that a patient should receive at any given time, and that this optimal dose... can change over time for each patient.” 

Conventionally, “treatment is often based on administering a maximum tolerated dose” with dose reductions based on toxicity, he notes. The study thus represents “a promising step forward for efficacy-guided dose adjustments” that also enables favorable treatment tolerability.  

‘Small Data’ Milestone 

Much of medical AI research involves retrospective analysis using pre-existing data from large populations to train a model to see if it would have effectively made a prediction, says Ho, such as whether a patient was going to be readmitted to the hospital or recovery would improve or worsen. “Real-world and prospective studies are less common, and many pertain to diagnosis.” 

Rarer still is the use of data to treat, he continues. “In our case, the dosage guidance was forward-looking [prospective] and the study was conducted in the real-world in that it was not using previous databases to look at treatment cases that concluded already.” 

It was in fact a milestone that CURATE.AI created clinically actionable patient profiles using “small data” to guide actual treatment, Ho adds. The platform is still being tested in multiple clinical trials, which will help determine how it will ultimately be deployed. 

Previously, CURATE.AI was used in a first-of-its-kind clinical trial to guide treatment for a patient with a rare blood cancer. Compared to the overall dose under the standard-of-care regimen, the trial’s recommended drug doses were lower and well-tolerated by the patient, who also saved about $8,000 over the first two years of treatment (npj Digital Medicine, DOI: 10.1038/s41746-024-01195-5). 

In earlier trials, CURATE.AI has also provided actionable “N-of-1” (single patient) combination therapy for the duration of patient care to achieve better outcomes for patients, with drug dosing optimized as patient responses were recorded, NUS has reported. These included a pilot clinical study conducted in collaboration with a U.S.-based hospital where a patient with advanced prostate cancer was recommended a 50% reduction in dose of an investigational inhibitor drug for increased efficacy. With the lower, more tolerable dose, the patient was able to resume an active lifestyle. 

Similarly, CURATE.AI recommended that a patient in Singapore with advanced cancer have the dose of a prescribed chemotherapy agent (nab-paclitaxel) reduced, after which his lung tumor shrank, and the cancer stopped progressing. The patient also stayed on the treatment longer than most patients given the drug. These findings led to the most recent feasibility study. 

Comprehensive Approach

The other trials underway involve indications beyond cancer, including a hypertension trial to assess the feasibility of a larger randomized controlled trial evaluating the efficacy of CURATE.AI-assisted dose titration intervention. The study protocol was published in European Heart Journal - Digital Health early last year (DOI: 10.1093/ehjdh/ztad063). 

An immunotherapy trial for CURATE.AI has already been completed and data analysis is underway, Ho says. Trials have also been wrapped up where the platform was used to “dose the difficulty of cognitive training platforms to optimize cognitive performance while also monitoring cognitive performance over time as well. This could be useful for broad communities of users, ranging from younger individuals looking to enhance cognitive function earlier in life, to older adults aiming to preserve or improve cognitive performance as they age.” 

A key differentiator of WisDM, notes Ho, is that it views technology deployment in a “highly comprehensive manner,” focusing on both technology validation in the clinic and user acceptability via extensive qualitative research. The users referenced here are patients and caregivers as well as the clinical community.  

WisDM also has expertise in behavioral economics, which is critical to overcoming adoption barriers, and has published extensively in this space. These include institutional review board/ethics board-driven research on topics ranging from factors driving the economic value of digital therapeutics (npj Digital Medicine, DOI: 10.1038/s41746-025-01600-7), physicians’ perceptions of integrating CURATE.AI into clinical practice (JMIR Human Factors, DOI: 10.2196/48476), and the attitudes and experiences of older adults toward using such digital therapeutic platforms (JMIR Formative Research, DOI: 10.2196/63568, and JMIR Human Factors, DOI: 10.2196/58641).  

Rethinking Trial Design

Next steps for Ho and his team include expanding the feasibility study into larger, randomized controlled trials to validate the effectiveness of the CURATE.AI platform against traditional treatment regimens. These will leverage the expertise of WisDM to assess factors such as savings to the patient and healthcare system (e.g., reduced readmissions due to toxicity), in addition to well-accepted efficacy and patient outcome measures, he says. 

“Our initial studies have also included feedback and oversight discussions with our clinical community to determine how we might assess for future patient recruitment, as well as additional recruitment sites we can explore, and possible international trial sites as well [currently in discussion],” says Ho. The research team is also considering bringing in industry partners with biomarker monitoring platforms able to provide longitudinal insights to assist with dose modulation. 

“Bringing the concept of small data and N-of-1 medicine to clinical trials was quite a journey,” Ho says. It required a multidisciplinary team spanning engineering and medicine, behavioral and health economics, implementation sciences, and trial design innovation, along with reimbursement considerations.  

AI for health and healthcare is “indeed promising,” and the task ahead is to see if it can help achieve prospective outcomes that help patients, he adds. “This [latest] study and others... illuminate the possibility of re-thinking how we design trials. We often view data as a snapshot [a single blood test or image] to guide care, and that care is often at a fixed dose.” 

But patients evolve over time, so it is “critical to evolve treatment alongside them and to adjust where necessary and possible,” says Ho. “In addition to helping patients, these concepts can also help to keep healthy people healthy or make them even healthier.”