A review of the electronic medical records of thousands of prostate cancer patients at two major medical institutions revealed a nearly two-fold increase in the rate of Alzheimer’s disease diagnosis among those treated with androgen deprivation therapy.
The study, by researchers at the Stanford University School of Medicine and the University of Pennsylvania Perelman School of Medicine, demonstrates emerging techniques for extracting biomedical data from ordinary patient medical records.
The paper was published online Dec. 7 in the Journal of Clinical Oncology. Nigam Shah, MBBS, PhD, associate professor of biomedical informatics research at Stanford, is the senior author. The lead author, Kevin Nead, MD, is a resident at the University of Pennsylvania who got his medical degree at Stanford.
Because testosterone can promote the growth of prostate tumors, clinicians have used androgen deprivation therapy to lower testosterone and other androgens in prostate cancer patients since the 1940s. In the United States, about a half-million men currently receive ADT as a treatment for prostate cancer.
The researchers scanned the records of about 5.5 million patients from two hospitals: Stanford Health Care, in Palo Alto, and, through a prior institutional research agreement, 3.7 million patients from Mount Sinai Hospital, in New York City. Among this cohort, they identified about 9,000 prostate patients at each institution, 16,888 of whom had non-metastatic prostate cancer. A total of 2,397 had been treated with androgen deprivation therapy.
Patients in the study who had been treated with ADT had about a 1.88 times increased rate of being diagnosed with Alzheimer’s disease in a median follow-up period of 2.7 years compared with prostate cancer patients who did not receive ADT, the study found. The subset of men treated with ADT for longer than 12 months had a 2.12 higher risk — more than double that of prostate cancer patients not treated with ADT.
Using existing clinical data
Shah said the idea for the study started with Nead, who noticed some references in the medical literature to men who had ADT treatment for prostate cancer subsequently experiencing cognitive declines. “There was some chatter in the literature,” said Shah. But no one had formally tried to find out if ADT therapy leads to cognitive defects.
“This is the kind of question that typically you would need a large clinical trial to answer,” said Shah. But a formal clinical trial would be enormously expensive. “So instead, we’re making secondary use of existing clinical data collected as part of routine medical care” — clinical data that’s practically free.
Although ADT may increase the risk of defects in cognition and hand-eye coordination for reasons other than Alzheimer’s disease, the team decided to focus specifically on Alzheimer’s because the condition is easier to identify in medical records, said Shah. “Broader dementias and vascular dementia are kind of hard to quantify and define, so we had to narrow the scope of the analysis to make it feasible with the methods that we have available,” he said.
After making statistical adjustments to control for biases, the team performed two kinds of tests to assess the reliability of the findings. In “falsification tests,” they looked for false associations in the data. Specifically, they looked for five associations with medical conditions such as tuberculosis and allergies unlikely to be connected to testosterone levels. Those tests all came back negative.
They also looked for associations likely to be positive, such as age and cardiovascular disease — both conditions known to be associated with a risk of Alzheimer’s disease. Those positive associations were confirmed by the data.
Patients who are receiving ADT and are concerned about the potential risks should discuss them with their physicians. “The association found in this study should be evaluated in the context of the overall treatment choices available to any specific patient,” Shah said. “This study demonstrates the value of using existing EMR data to quantify the trade-offs that various treatments offer.”
Stanford well-situated for work
Although other institutions are beginning to use patient records to ask and answer research questions, Stanford is unusually well-situated to do this work, said Shah. Patient records at Stanford are managed by the Stanford Center for Clinical Informatics. “If I was to go to another institution and ask for this same data, I’d probably be waiting a year, a year and a half; there would be so much hassle involved in being able to access the clinical documents that have detailed patient health information in them, whereas here we have the necessary infrastructure in place so that once you get Institutional Review Board approval, getting to the data doesn’t take you a year and a half or two years,” he said. Depending on the complexity of the data, it could take as little as two weeks at Stanford. Worst case, it could take six months, he said. “But it’s not in geological time-scale.”
The work was bolstered by electronic medical records shared by Mount Sinai Hospital, which doubled the number of relevant patient records — highlighting the importance of such cross-institutional collaborations. Harnessing the data in electronic medical records is part of Stanford Medicine’s efforts in precision health — health care whose goal is to anticipate and prevent disease in the healthy and precisely diagnose and treat disease in the ill.
Other Stanford-affiliated co-authors of the paper are medical student Gregory Gaskin; life science research assistant Cariad Chester; and associate professor of surgery and of medicine Nicholas Leeper, MD. The researchers collaborated with Joel Dudley, PhD, assistant professor of genetics and genomic sciences at the Icahn School of Medicine at Mount Sinai.
This research was supported by the National Institutes of Health (grant U54HG004028 for the National Center for Biomedical Ontology), the National Library of Medicine (grant R01LM011369) and the National Institute of General Medical Sciences (grant R01GM101430).
Shah is an inventor on patents owned by Stanford University that enable the use of clinical text for data-mining.