Sleep Disorder and Memory Problems May Signal Parkinson’s Years Before Diagnosis

Long before the tremors and stiffness of Parkinson’s disease announce themselves, the body often sends quieter signals — disturbed sleep, slipping memory, lightheadedness on standing, and stubborn constipation. A new retrospective study drawing on one of the largest and most diverse biomedical databases in the United States has now mapped which of these early, non-motor symptoms are most strongly linked to a later Parkinson’s diagnosis, and the results sharpen a long-standing question in neurology: how early can the disease be detected, and how reliably?

The research, published in npj Parkinson’s Disease, was conducted by a team led by Meet Popatbhai Kachhadia of Florida Atlantic University’s Charles E. Schmidt College of Medicine and the Marcus Neuroscience Institute, together with colleagues at Henry Ford Hospital. The investigators mined the All of Us Registered Tier dataset, a National Institutes of Health program that aggregates electronic health records from hundreds of thousands of volunteers across the country. Their analysis, built on the Curated Data Repository version R2025Q4R6, compared people who went on to develop Parkinson’s disease with carefully matched individuals who did not, tracing the medical footprints both groups left in the years preceding diagnosis.

The design was deliberately rigorous. To qualify as a case, a participant needed at least two distinct Parkinson’s disease diagnosis dates recorded in their health record and had to be at least 40 years old, a threshold intended to filter out coding errors and incidental mentions. Each case was then matched to four controls — people without Parkinson’s — who shared the same sex and birth year. This matching strategy, known as a matched case-control design, allows researchers to isolate the effect of the symptoms under study while holding constant some of the demographic factors that could otherwise muddy the comparison. In total, the cohort comprised 2,673 cases and 10,692 matched controls, with a mean age of 66.8 years and 61.2 percent male participants.

The symptoms of interest were nine prodromal non-motor features: rapid eye movement sleep behavior disorder, mild cognitive impairment, orthostatic hypotension, constipation, urinary dysfunction, depression, anxiety, insomnia, and erectile dysfunction. Identifying these conditions in electronic health records is harder than it sounds, because the same clinical reality can be documented under many different codes. The team therefore generated their feature definitions through a name-based discovery process over standard SNOMED condition concepts, expanding each concept to include all of its descendants in the SNOMED hierarchy and then auditing every concept individually. The full concept sets are reported in the paper, an unusual degree of transparency that other researchers can now reuse and scrutinize.

The statistical centerpiece was conditional logistic regression, a method tailored to matched designs that estimates the odds of each symptom appearing before the index date in cases relative to their matched controls. The results were striking in their hierarchy. Rapid eye movement sleep behavior disorder — a condition in which people physically act out their dreams, sometimes violently — showed by far the strongest association with subsequent Parkinson’s disease, with an odds ratio of 9.68 and a 95 percent confidence interval of 5.89 to 15.90. In practical terms, people with documented REM sleep behavior disorder were nearly ten times more likely to be diagnosed with Parkinson’s than their matched peers, a magnitude consistent with the idea that the brainstem circuits governing dream paralysis are among the earliest casualties of the disease process.

Mild cognitive impairment came next, with an odds ratio of 4.10 (95 percent CI 3.09 to 5.46), followed by orthostatic hypotension — an abnormal drop in blood pressure upon standing — at 1.83 (95 percent CI 1.46 to 2.30). More modest but still statistically reliable associations emerged for constipation (odds ratio 1.30), urinary dysfunction (1.27), depression (1.23), and anxiety (1.19). Insomnia, by contrast, showed no association at all, with an odds ratio of 0.98, suggesting that ordinary difficulty sleeping is not part of the prodromal Parkinson’s signature in health record data. Erectile dysfunction displayed a statistically significant inverse association — an odds ratio of 0.83 — but the authors were careful to interpret this as an artifact of differential ascertainment, meaning the condition was likely recorded differently in the two groups, rather than as evidence of any protective effect. The estimate was unstable across sensitivity analyses and should not be taken at face value.

Robustness was tested through ten separate sensitivity analyses, including three alternative definitions of mild cognitive impairment and a restriction of the analysis to cases who had been treated with levodopa, the standard Parkinson’s medication. The estimates for REM sleep behavior disorder, mild cognitive impairment, and orthostatic hypotension held steady across all of these checks, lending confidence that they reflect genuine prodromal associations rather than coding quirks. Depression and anxiety, however, attenuated to the null when the analysis was restricted to levodopa-treated cases, hinting that their apparent links to Parkinson’s may partly reflect shared documentation patterns or the presence of other conditions rather than a direct prodromal relationship.

Perhaps the most sobering finding concerns prediction at the level of the individual patient. Despite the strong group-level associations, the overall discrimination of the symptom profile was modest, with a concordance statistic of 0.58 and an area under the curve of 0.585 — barely better than a coin flip. In other words, knowing that a person has one or more of these documented non-motor symptoms tells you something meaningful about populations, but it does not allow a clinician to say with confidence that any particular individual will develop Parkinson’s. The authors are explicit on this point: electronic health record-documented non-motor features are informative at the group level but limited as individual-level predictors. This distinction matters enormously for how such findings are communicated to the public, because a tenfold odds ratio can sound alarming even when the absolute risk for any one person remains small.

The study also carries a methodological footnote that reflects the changing face of scientific research: the authors disclose that generative artificial intelligence, specifically Claude from Anthropic, was used to assist with the development and review of statistical analysis code and with drafting and editing of the manuscript. Crucially, they state that all study design decisions, data extraction, statistical analyses, verification of results, and interpretation were performed by the authors themselves, who take full responsibility for the content. No dedicated funding was received for the research, and the authors declare no competing interests. The work relied entirely on data contributed by All of Us participants, whose willingness to share their health records made the analysis possible.

What emerges from this study is a nuanced picture of Parkinson’s long shadow. The disease’s prodromal phase — the years during which pathology silently spreads through the nervous system before motor symptoms appear — leaves traces in the medical record, and the strongest of those traces involve the sleep disorder in which dreams break loose from their usual paralysis, early cognitive slowing, and unstable blood pressure. These findings align with a growing body of evidence that Parkinson’s is not merely a disease of movement but a systemic neurodegenerative process with a long runway. For researchers, the well-documented concept sets and the transparent sensitivity analyses offer a reusable template for mining large biobanks for early signals. For clinicians, the message is more cautious: these symptoms are worth noting, particularly in combination, but the era of reliably predicting Parkinson’s in an individual from health record data alone has not yet arrived. The next step, the study implicitly suggests, lies in combining these non-motor trajectories with emerging biomarkers — from smell testing to imaging and fluid-based markers — to build prediction tools that work not just for populations but for the person sitting in the examination room.

Subject of Research: Prodromal non-motor symptom trajectories preceding Parkinson disease diagnosis identified through retrospective analysis of electronic health records

Article Title: Prodromal non-motor symptom trajectories preceding Parkinson disease diagnosis: a retrospective matched case-control study using the All of Us Research Program

Article References: Kachhadia, M. P., Husain Wilson, S., Swerdloff, M. A., & Shaikh, J. D. (2026). Prodromal non-motor symptom trajectories preceding Parkinson disease diagnosis: a retrospective matched case-control study using the All of Us Research Program. npj Parkinson's Disease. https://doi.org/10.1038/s41531-026-01591-6

Image Credits: AI Generated

DOI: 10.1038/s41531-026-01591-6

Keywords: Parkinson's disease, prodromal symptoms, REM sleep behavior disorder, mild cognitive impairment, orthostatic hypotension, All of Us Research Program, electronic health records, case-control study, non-motor symptoms, neurodegeneration, early detection, conditional logistic regression