Natural Language Processing Algorithm Evaluates Trials’ Kidney Function Inclusion Criteria

Investigating whether phase III clinical trial kidney function eligibility requirements are appropriate, Dr. Kristian Stensland, MPH, a research fellow in the Department of Urology at the University of Michigan, and collaborators developed a natural language processing (NLP) method to analyze study inclusion criteria.

Dr. Stensland, presenting findings at the 22nd Annual Scientific Meeting in Urologic Oncology, reported that a third of phase III urologic oncology trials had kidney function criteria, and 14% excluded patients on the basis of kidney function alone. Notably, only about half of those trials tested interventions with potential renal clearance or toxicity. Conversely, some trials didn’t have any eligibility requirements despite the study therapy having potential effects on kidney function.

The researchers used NLP to pull kidney function requirements from registered phase III urologic oncology clinical trials. The results were 90% accurate with one kidney function miss by the algorithm and four minor disagreements in the level of glomerular filtration rate restriction.

Among the 850 trials comprised of 1,116 interventions, 297 (34%) listed kidney function eligibility restrictions and 421 (50%) tested an intervention with potential kidney effects. Of the 120 trials with the strictest exclusions (GFR<45 or GFR<60), 43% tested interventions with no significant renal effects. Conversely, of 421 trials testing interventions with potential renal toxicity or significant renal clearance, only 169 (37%) had kidney function exclusions in the eligibility criteria.

Dr. Stensland posited that their NLP algorithm “allows for scalable, fast, and reliable characterization of clinical trial kidney function eligibility criteria, and is a proof of concept for a platform to analyze additional eligibility criteria of clinical trials on a broader scale.”

Given the disparities found in the study, he also called for further examinations of enrollment criteria and the extent to which they impact enrollment in trials are warranted. “Additional work is needed to further describe existing trial eligibility criteria and the potential effects on both trial enrollment and the generalizability of expanding eligibility criteria,” he closed.