Abstract
Study Objectives:
Individuals with isolated rapid eye movement sleep behavior disorder (iRBD) are at high risk of neurodegenerative parkinsonian disorders or dementia (NPD). Determining which characteristics predict greatest risk could improve clinical care. Our objectives were to utilize electronic health record data to apply prodromal Parkinson’s disease research diagnostic criteria to outpatients with iRBD and determine their utility for identifying patients with iRBD at high vs low risk for NPD diagnosis.
Methods:
This was a retrospective cohort study at a tertiary care center in western Pennsylvania. Diagnosis of iRBD was confirmed with expert manual chart review. Prodromal risk markers and signs and symptoms were determined with diagnostic codes. Multivariable Cox proportional hazards models examined a range of covariates as predictors of time to NPD diagnosis.
Results:
Of 448 patients with iRBD, 82 (18.30%) were diagnosed with NPD. Forty-nine (10.93%) had > 80% prodromal Parkinson’s disease probability. There was no difference in time to NPD among those who met vs did not meet > 80% probability (log-rank P = .49). In a Cox model that included all assessed criteria features, risk of diagnosis was associated with male sex (hazard ratio = 2.06, 95% confidence interval = 1.04–1.10), older baseline age (hazard ratio = 1.07, 95% confidence interval = 1.05–1.10), and cognitive dysfunction diagnostic code (hazard ratio = 2.83, 95% confidence interval = 1.79–4.46). Time to NPD diagnosis among patients with predicted high vs low risk was significantly different (log-rank test P = .012).
Conclusions:
In outpatients with iRBD, a model combining individual Parkinson’s disease risk factors and prodromal features accurately identifies individuals at high risk for NPD diagnosis. Results demonstrate the potential of electronic health record data to translate research on prodromal Parkinson’s disease to the clinic.
Citation:
Chahine LM, Newman A, Boyce RD, Brooks MM. Prodromal features and risk of neurodegenerative disorders diagnosis in outpatients with REM sleep behavior disorder. J Clin Sleep Med. 2025;21(11):1857–1869.
Keywords: REM sleep behavior disorder, Parkinson’s disease, risk assessment, electronic health record data
BRIEF SUMMARY
Current Knowledge/Study Rationale: Determining which characteristics predict the greatest risk for neurodegenerative disorder diagnosis among individuals with isolated rapid eye movement sleep behavior disorder could improve clinical care. We determined the utility of prodromal Parkinson’s disease research diagnostic criteria for identifying individuals at high risk for diagnosis in the real-world outpatient setting.
Study Impact: In outpatients with isolated rapid eye movement sleep behavior disorder, prodromal Parkinson’s disease diagnosis based on research criteria did not effectively discriminate between individuals with isolated rapid eye movement sleep behavior disorder who were at high vs low risk for neurodegenerative parkinsonian disorders or dementia diagnosis; however, a model combining risk factors and prodromal features, especially age, sex, and cognition, improved prediction. Results demonstrate the potential of electronic health record data to translate research on prodromal Parkinson’s disease to the clinic.
INTRODUCTION
Neurodegenerative parkinsonian disorders or dementia (NPD), including Parkinson’s disease (PD), dementia with Lewy bodies, and multiple system atrophy are among the most common and disabling diseases of older adulthood, and their prevalence is increasing.1,2 At the time of NPD diagnosis, advanced neuronal loss has already occurred.3 In order to prevent disease progression to disability, NPD must be diagnosed during the earliest stages of neuropathology, before the full clinical syndromes that constitute diagnostic criteria for these diseases emerge.4
Because of their well-recognized neuropathology and progression, patients with NPD are prime candidates for improving early detection during the prodrome of these diseases.4 Several characteristics of the prodrome have been described.5,6 Many prodromal signs and symptoms are nonspecific nonmotor features, such as constipation and depression, which are common in older adults who do not develop NPD. Other features are more specific. In particular, the sleep disorder rapid eye movement sleep behavior disorder (RBD) is one of the early manifestations that carries high risk for underlying alpha-synuclein pathology.7 Isolated RBD (iRBD) is a parasomnia marked by increased muscle tone during rapid eye movement sleep and manifests with dream enactment behavior in the absence of identifiable secondary etiologies.8 The disorder is strongly predictive of future risk of NPD, with over half of individuals with iRBD enrolled in research studies being diagnosed with one of these diseases over 5 years.9 However, there are many individuals with iRBD who do not go on to develop parkinsonism or dementia over long periods of follow-up, despite having multiple prodromal features.10 Accurately identifying individuals with iRBD who are at high short-term risk of clinical progression is of great interest since this population has been targeted for testing interventions to prevent disability from NPD.11 This information could also inform the clinical care of these individuals, including prognostic counseling,12 early referral to specialists, and early diagnosis.
In vivo biomarkers have been developed and validated that accurately identify which individuals with iRBD or other prodromal features go on to receive a diagnosis of PD or related disorders in the research setting.9 At present, invasive procedures, such as cerebrospinal fluid acquisition, or costly imaging tests, such as dopamine transporter single-photon emission computerized tomography scan, are required to assess the most widely studied biomarkers.5,6,13 These biomarkers are not validated yet for clinical use.
Based on data from longitudinal cohort studies, criteria have been put forth by the Movement Disorders Society (MDS)5,6 that combine signs, symptoms, and biomarkers to determine a probability for prodromal PD based on any available assessments. The prodromal features that contribute most substantially to the overall probability are polysomnographically confirmed RBD, dopaminergic imaging abnormalities, and genetic features. In the clinic, where biomarkers are not routinely acquired and where prodromal features may or may not be systematically ascertained during routine clinical care, the utility of the MDS prodromal criteria or other methods to accurately identify risk of NPD diagnosis among individuals with iRBD is not well understood. This is of particular relevance for the approximately 40% of outpatients with iRBD diagnosis for whom polysomnographic confirmation is not available.12,14,15
Studies examining diagnostic codes and presenting symptoms as discrete variables have indicated that several prodromal symptoms are seen in individuals receiving outpatient care who go on to receive a diagnosis of PD.16,17 However, gaps in knowledge remain as to how to accurately determine risk among patients with iRBD seen in the clinic. Identifying features assessed during routine clinical care that alone or in combination predict high-risk individuals with iRBD could translate into clinical support tools built into the electronic health record (EHR) that combine EHR features to efficiently synthesize large amounts of data into useful information for the provider.
In this analysis, we utilized EHR data to study a real-world, clinic-based population of individuals with iRBD. We had 2 primary objectives: (1) compare prevalence of prodromal features in those who go on to receive an NPD diagnosis vs those who do not, and (2) determine the ability of the research criteria for prodromal PD to identify individuals with iRBD at high vs low risk for NPD diagnosis. Secondary objectives were to (1) explore incorporation of a wider range of prodromal features using EHR data to develop models to predict a high risk for NPD and (2) determine the utility of diagnostic codes to identify the onset of clinical diagnosis of NPD (“phenoconversion”).
METHODS
This was a retrospective cohort study of individuals with iRBD. This study was approved by the University of Pittsburgh Institutional Review Board with waiver of informed consent.
Sample
The sample consisted of individuals receiving care at a tertiary care center in western Pennsylvania. As previously described,14 a query of the EHR was conducted to identify individuals who were assigned an International Classification of Diseases (ICD) version 9 or 10 code for RBD between January 1, 2011, and May 31, 2021. Manual chart review was performed as previously described14 to determine which patients among those individuals had possible, probable, or definite iRBD. Level of certainty of iRBD was designated as definite when rapid eye movement sleep without atonia was demonstrated on polysomnogram (PSG), probable when PSG was not available but the EHR indicated consistent health care delivery for suspected iRBD (treatment initiation for dream enactment behavior and response to treatment), and possible when the aforementioned criteria were not met but the EHR indicated presence of dream enactment behavior possibly consistent with iRBD and no other explanation was present.
Assessment of risk markers and prodromal features using ICD codes
Risk markers and established or putative prodromal signs and symptoms were determined with ICD-9 and ICD-10 codes. Only the presence of an ICD-9 or ICD-10 code at and before the RBD ICD code date was considered evidence of a sign, symptom, or risk factor.
Established signs and symptoms were those that have sufficient evidence as determined by MDS prodromal criteria or recent meta-analyses: olfactory loss, constipation, urinary dysfunction, sexual dysfunction, depression, cognitive dysfunction, hypersomnolence, orthostatic hypotension, and RBD. Putative signs or symptoms were those reported in the literature to be possible prodromal features.18,19
Established risk markers for PD5,6 included in the MDS prodromal criteria were diabetes, hypouricemia, nonuse of caffeine, nonuse of nicotine, pesticide exposure, and occupational solvent exposure, whereas hyperuricemia, nicotine use, and caffeine use are protective. Given the extremely low expected frequency of use of ICD codes to represent caffeine, pesticide, and organic solvent exposure (which was confirmed subsequently in our dataset), these were not included in our analysis.
Calculation of probability for prodromal PD based on MDS criteria
The presence of established risk markers and signs/symptoms were utilized to apply the MDS criteria,5,6 in which probability of prodromal PD is determined after calculation of risk based on a Bayesian naïve classifier.
A given individual’s baseline pretest (“prior”) probability of a disease was determined based on their age, and pretest odds were subsequently calculated. The MDS criteria provide a prior probability for those aged 50 years and older. For our analysis, pretest probability for those aged 40 to younger than 50 years was specified as 0.2%, and for all those younger than 40 years, it was specified as 0.001% based on limited available data on incidence of young-onset PD and the very limited data on epidemiology of juvenile parkinsonism.20
Information on signs, symptoms, and risk markers from the EHR was then utilized to determine a new (“posttest”) probability. The MDS criteria designate likelihood ratios (LRs), which are based on the strength of associations between each feature and incident PD risk based on existing evidence in the literature at the time the criteria were published. LRs were assigned according to data available in the EHR at and before the RBD ICD code date. Positive LR (LR+) or negative LR (LR−) was assigned when an ICD code or other data point allowed for this designation; LR− was only assigned when absence of a given feature was documented. Otherwise, an LR of 1 was assigned.
For each participant, LRs for each assessed feature were multiplied to compute the total LR:
To determine the utility of the MDS prodromal criteria when PSG confirmation of iRBD is present, and when it is not available, 2 total LRs were computed: 1 where the LR for PSG-proven RBD (LR = 130) was assigned to the definite RBD group, and 1 where LR = 2.8 was assigned to all in the iRBD group, hereafter referred to as total LRPSG+ and total LRPSG−, respectively.
Posttest odds were computed by combining pretest probability with each total LR (pretest odds × total LR). Subsequently, probability for prodromal PD was determined:
A probability of > 80% is consistent with high probability and is sensitive for incident PD diagnosis in the research setting.21 Two other thresholds were examined: 50% and 30% probability of prodromal state. These probabilities were examined when accounting for PSG confirmation (total LRPSG+) and when PSG was not factored into the calculation (total LRPSG−).
Incorporating non-ICD code EHR data into MDS criteria determination
Discrete data in the EHR besides ICD codes that were available through the health records query system were reviewed to identify additional possible sources of information on prodromal risk markers or symptoms. Additional sources of data were identified for 4 risk markers: family history, smoking history, hyperuricemia, and diabetes. These additional data were used to define the presence of risk markers as follows.
(1) Family history: considered to be present when a family history of PD in a first- or second-degree family member was entered into the EHR tab for family history. Because pertinent negatives for PD might not be entered as discrete data in the EHR, absence of data could not be assumed to be absence of family history.
(2) Smoking history: data from the social history tab of the EHR, which includes a smoking subsection, were analyzed. This section in the EHR is often, although not always, updated when a patient has an office visit encounter. A patient was designated as a current cigarette smoker when data indicated smoking of any amount at or within 1 year prior to the RBD ICD code date, a former smoker when data indicated at least 1 pack/year of smoking but the patient had ceased smoking at the time of ICD code date, and a never smoker when the smoking history had been updated within 1 year prior to or any time after the ICD code date and the patient was documented to have never smoked.
(3) Diabetes: laboratory tests for glycated hemoglobin (HbA1c) were used to define diabetes. The HbA1c represents an average of glycemia over the 3 prior months. An HbA1c ≥ 6.5% at or before the RBD ICD code date was considered consistent with diabetes. An HbA1c < 5.6% within 3 months of the RBD code date was considered an absence of diabetes.
(4) Hyperuricemia: laboratory-assessed uric acid at any time before the RBD ICD code date. A patient was designated as “ever” having had high urate if they ever had a uric acid level of > 5.6 units at any time prior to the ICD code date.
We then combined ICD code and laboratory tests for diabetes/HbA1c, gout/hyperuricemia, and nicotine/smoking history and applied LR+ or LR− accordingly. A modified total LR was then generated as follows, incorporating the additional information from the EHR (total LRICD+EHR):
Study endpoints
The primary outcome in this analysis was time to clinical diagnosis of NPD where the date of iRBD diagnosis was designated as time 0. This was determined through manual chart review of all patients with iRBD by an expert movement disorder neurologist (author LC), as described previously.14 For those who did not receive an NPD diagnosis, the last follow-up was defined as the date of the last informative encounter in the chart or date of death.
To explore whether EHR data can be useful in identifying phenoconversion in iRBD, we also explored the relationship between NPD diagnosis as determined by expert manual review and representation of NPD diagnosis in ICD codes.
Statistical analysis
The baseline visit for this analysis was the date on which an encounter was first associated with an iRBD ICD code. The primary outcome was a diagnosis of NPD as ascertained with manual chart review. The primary predictors were the MDS Prodromal PD criteria, examined separately as 3 different representations: (1) prodromal probability, (2) meeting different cutoffs of prodromal probability, and (3) individual prodromal features and risk markers.
Baseline characteristics of the sample were examined and compared in those who did vs did not receive NPD diagnosis. Sample characteristics were summarized with descriptive statistics. Differences in demographics, risk markers, and prodromal features were described. Two-sample t test or Wilcoxon rank-sum test for continuous variables and chi-square or Fisher’s exact test for categorical variables were applied as appropriate to compare groups.
The log-rank test compared differences in time to NPD among those who met 80%, 50%, and 30% prodromal threshold vs those who did not, when applying MDS criteria based on total LRPSG+ and total LRPSG−.
To determine whether individuals with iRBD at high risk for NPD can be identified in our EHR dataset, 2 models were trained and compared to each other: 1 that included individual MDS criteria risk markers (CoxMDS) as covariates and 1 that included features selected by least absolute shrinkage and selection operator (LASSO) (CoxLASSO) as covariates (see below). The sample was randomly split into 80% training and 20% test samples, with 20% in each of the training and test samples diagnosed with NPD.
To examine each individual MDS criteria risk marker and prodromal features present at the time of RBD ICD code as predictors of NPD diagnosis, a multivariable Cox proportional hazards model was conducted on the train dataset. The outcome in the model was NPD; individuals who never demonstrated evidence for NPD were censored at death or last follow-up. Covariates included individual components of the MDS criteria: age at RBD, sex, smoking, diabetes, gout, RBD level of certainty, olfaction, constipation, hypersomnolence, orthostasis, erectile dysfunction, urinary dysfunction, depression, and cognition. Tied survival times were handled using the Breslow approximation. Statistical significance was set at P < .05. Based on the median predicted relative-hazard ratio in the train dataset, patients were classified as high risk (above the median) or low risk (below the median) for NPD in the test dataset. Log-rank test and Kaplan-Meier plot were utilized to compare NPD diagnosis-free survival in the predicted low- and high-risk groups in the test dataset.
To develop an exploratory predictive model using a range of possible prodromal features, a Cox LASSO model was conducted on the train dataset. The LASSO model was selected given its suitability in the setting of high-dimensional data where the number of possible predictors is much larger than the number of outcomes observed. The objective was to extract features that distinguish those who were short-term vs long-term survivors without NPD. A LASSO Cox model was trained, and features were selected on the training set. Regularization parameters were selected by 10-fold cross-validation on the training set. Tied survival times were handled using the Breslow approximation. A multivariate Cox proportional hazards model that included only variables selected by LASSO was then conducted.
To compare the CoxMDS to the CoxLASSO, Harrell’s C and Somers’ D were computed. A value of 0.5 for Harrell’s C indicates no predictive discrimination, and values of 0 or 1.0 indicate perfect separation of patients with different outcomes. Somers’ D estimates the probability of a survivor having a lower hazard ratio than a nonsurvivor.
To compare methods of identifying prodromal features, we examined the number of individuals who had concordance between variables that were ascertained via different methods: (1) smoking history and ICD code for nicotine abuse, (2) gout and serum urate, and (3) diabetes and HbA1c. The proportion of individuals who would be classified as having a risk factor or prodromal feature using ICD code alone vs using additional EHR data was examined, as were those who met MDS prodromal threshold based on the total LRICD+EHR.
Concordance between NPD diagnosis based on expert manual chart review (clinically defined “phenoconversion”) and ICD code for NPD (ICD-defined “phenoconversion”) was examined. Sensitivity and specificity of ICD code for NPD compared to manual chart review was calculated. All analyses were conducted with Stata version 18 (StataCorp, College Station, Texas).
RESULTS
There were 460 individuals who received an RBD ICD code over the search period and who were determined after manual chart review to have iRBD. Twelve did not have any follow-up after baseline and were excluded from analysis.
Of the remaining 448 individuals, 82 (18%) were diagnosed with an NPD, as confirmed by expert manual chart review. Median time from RBD ICD code to diagnosis was 3.78 (25–75%; interquartile range, 1.80–5.95) years. Among those who were not diagnosed with NPD, time from ICD code to last follow-up in the EHR was 4.59 (25–75%; interquartile range, 2.38–6.96) years.
Among the 82 diagnosed with NPD, diagnoses were distributed as follows: 33 (40.24%) PD, 12 (14.63%) dementia with Lewy bodies, 5 (6.10%) multiple system atrophy, and 32 (39.02%) other dementia. Sample characteristics are shown in Table 1 in the group as a whole and stratified by those diagnosed with NPD vs those who were not. Those who were diagnosed were significantly older at baseline RBD code compared to those who were not (mean (SD) age at baseline 69.34 [8.71] vs 57.26 [15.73] years, respectively; P < .001).
Table 1.
Sample characteristics of patients with iRBD at baseline and likelihood ratios assigned stratified by subsequent diagnosis of NPD and in the train and test dataset.
| Total | Not Diagnosed With NPD | Diagnosed With NPD | P Value | ||
|---|---|---|---|---|---|
| n = 448 | n = 366 | n = 82 | |||
| Age at first RBD code | 59.47 (15.42) | 57.26 (15.73) | 69.34 (8.71) | <.001 | |
| Race | White | 417 (93.08) | 340 (92.90) | 77 (93.90) | .84 |
| Black | 19 (4.24) | 15 (4.10) | 4 (4.88) | ||
| AAPI | 6 (1.34) | 5 (1.37) | 1 (1.22) | ||
| Missing/declined | 6 (1.34) | 6 (1.64) | 0 (0.00) | ||
| Sex at birth | Female (LR = 0.8) | 145 (32.4) | 124 (33.9) | 21 (25.6) | .15 |
| Male (LR = 1.2) | 303 (67.6) | 242 (66.1) | 61 (74.4) | ||
| Nicotine use | No nicotine abuse (LR = 1) | 244 (54.5) | 204 (55.7) | 40 (48.8) | .25 |
| Ever nicotine abuse (LR = 0.91) | 204 (45.5) | 162 (44.3) | 42 (51.2) | ||
| Diabetes | No data (LR = 1) | 320 (71.4) | 268 (73.2) | 52 (63.4) | .076 |
| Yes (LR = 1.5) | 128 (28.6) | 98 (26.8) | 30 (36.6) | ||
| Gout | No or female (LR = 1) | 422 (94.2) | 346 (94.5) | 76 (92.7) | .52 |
| Yes in male (LR = 1) | 26 (5.8) | 20 (5.5) | 6 (7.3) | ||
| RBD | Possible or probable (LR = 2.8) | 385 (85.9) | 316 (86.3) | 69 (84.1) | .61 |
| PSG proven (LR = 130) | 63 (14.1) | 50 (13.7) | 13 (15.9) | ||
| Olfactory dysfunction | No data (LR = 1) | 436 (97.3) | 359 (98.1) | 77 (93.9) | .034 |
| Yes (LR = 6.4) | 12 (2.7) | 7 (1.9) | 5 (6.1) | ||
| Constipation | No data (LR = 1) | 351 (78.3) | 292 (79.8) | 59 (72.0) | .12 |
| Yes (LR = 2.5) | 97 (21.7) | 74 (20.2) | 23 (28.0) | ||
| Hypersomnolence | No data (LR = 1) | 405 (90.4) | 336 (91.8) | 69 (84.1) | .033 |
| Yes (LR = 2.7) | 43 (9.6) | 30 (8.2) | 13 (15.9) | ||
| Orthostasis | No data (LR = 1) | 413 (92.2) | 337 (92.1) | 76 (92.7) | .85 |
| Yes (LR = 3.2) | 35 (7.8) | 29 (7.9) | 6 (7.3) | ||
| Sexual dysfunction | No data or female (LR = 1) | 376 (83.9) | 307 (83.9) | 69 (84.1) | .95 |
| Yes (LR = 3.4) | 72 (16.1) | 59 (16.1) | 13 (15.9) | ||
| Urinary dysfunction | No data (LR = 1) | 281 (62.7) | 239 (65.3) | 42 (51.2) | .017 |
| Yes (LR = 2.0) | 167 (37.3) | 127 (34.7) | 40 (48.8) | ||
| Depression | No data (LR = 1) | 249 (55.6) | 198 (54.1) | 51 (62.2) | .18 |
| Yes (LR = 1.6) | 199 (44.4) | 168 (45.9) | 31 (37.8) | ||
| Cognitive impairment | No data (LR = 1) | 332 (74.1) | 293 (80.1) | 39 (47.6) | <.001 |
| Yes (LR = 1.8) | 116 (25.9) | 73 (19.9) | 43 (52.4) | ||
| Estimated posterior probability for prodromal PD, total LRPSG+ calculation | 12.0 (2.2–40.0) | 9.4 (1.3–34.3) | 32.5 (14.2–60.5) | <.001 | |
| Prodromal probability >80%, total LRPSG+ calculation | No | 399 (89.1) | 331 (90.4) | 68 (82.9) | .049 |
| Yes | 49 (10.9) | 35 (9.6) | 14 (17.1) | ||
| Prodromal probability >50%, total LRPSG+ calculation | No | 355 (79.2) | 298 (81.4) | 57 (69.5) | .016 |
| Yes | 93 (20.8) | 68 (18.6) | 25 (30.5) | ||
| Prodromal probability >30, total LRPSG+ calculation | No | 303 (67.6) | 265 (72.4) | 38 (46.3) | <.001 |
| Yes | 145 (32.4) | 101 (27.6) | 44 (53.7) | ||
| Estimated posterior probability for prodromal PD, total LRPSG− calculation | 9.5 (1.9–27.7) | 6.5 (1.2–20.3) | 27.7 (11.0–48.1) | <.001 | |
| Prodromal probability >80%, total LRPSG− calculation | No | 438 (97.8) | 358 (97.8) | 80 (97.6) | .89 |
| Yes | 10 (2.2) | 8 (2.2) | 2 (2.4) | ||
| Prodromal probability >50%, total LRPSG− calculation | No | 397 (88.6) | 332 (90.7) | 65 (79.3) | .003 |
| Yes | 51 (11.4) | 34 (9.3) | 17 (20.7) | ||
| Prodromal probability >30%, total LRPSG− calculation | No | 342 (76.3) | 297 (81.1) | 45 (54.9) | <.001 |
| Yes | 106 (23.7) | 69 (18.9) | 37 (45.1) | ||
AAPI = Asian American and Pacific Islander, iRBD = isolated rapid eye movement sleep behavior disorder, LR = likelihood ratio, NPD = neurodegenerative parkinsonian disorders or dementia, PD = Parkinson’s disease, PSG = polysomnography, RBD = rapid eye movement sleep behavior disorder.
The prevalence of risk factors and prodromal features as defined by MDS criteria are shown in Table 1. Established prodromal symptoms were common, with depression present in 44.4% of the sample, urinary dysfunction in 37.3%, and cognitive impairment in 25.9%.
When probability for prodromal PD was calculated utilizing total LRPSG+, in the group as a whole, 49 (10.5%) had > 80% probability for prodromal PD. Most individuals who met the 80% threshold were in the definite iRBD group, and only a minority were in the probable or possible iRBD group (39 [79.59%], 6 [12.24%], and 4 [8.16%], respectively; P < .001), as expected, considering the much higher LR assigned for definite (PSG-confirmed) iRBD. When we recalculated probability for prodromal PD criteria with total LRPSG−, only 10 (2.2%) met 80% threshold.
As for other putative risk markers and prodromal sign and symptoms that are represented in ICD codes (Table 2), fatigue and anxiety were common nonmotor symptoms, occurring in half the sample. In terms of motor symptoms, tremor was present in 11.8% and gait abnormality in 22.8%. Hypertension and dyslipidemia were common risk factors, in over two-thirds.
Table 2.
Prevalence of putative risk factors or prodromal features in the sample as a whole, stratified by subsequent diagnosis of NPD.
| ICD Code at/Before RBD Code | Total | No | Yes | P Value* |
|---|---|---|---|---|
| n = 448 | n = 366 | n = 82 | ||
| Risk factors | ||||
| Hypertension | 294 (65.6) | 234 (63.9) | 60 (73.2) | .11 |
| Dyslipidemia | 303 (67.6) | 235 (64.2) | 68 (82.9) | .01 |
| Head injury | 45 (10.0) | 38 (10.4) | 7 (8.5) | .62 |
| Alcohol disorder | 23 (5.1) | 18 (4.9) | 5 (6.1) | .66 |
| Motor and musculoskeletal | ||||
| Tremor | 53 (11.8) | 37 (10.1) | 16 (19.5) | .02 |
| Gait abnormality | 102 (22.8) | 76 (20.8) | 26 (31.7) | .03 |
| Stiffness of joint | 13 (2.9) | 12 (3.3) | 1 (1.2) | .32 |
| Shoulder problem | 100 (22.3) | 81 (22.1) | 19 (23.2) | .84 |
| Osteoarthritis | 232 (51.8) | 184 (50.3) | 48 (58.5) | .18 |
| Neuropsychiatric | ||||
| Anxiety/panic disorders | 220 (49.1) | 185 (50.5) | 35 (42.7) | .20 |
| Bipolar disorder | 32 (7.1) | 30 (8.2) | 2 (2.4) | .067 |
| Sleep/fatigue | ||||
| Fatigue | 241 (53.8) | 190 (51.9) | 51 (62.2) | .09 |
| Sleep disturbance NOS | 81 (18.1) | 68 (18.6) | 13 (15.9) | .56 |
| Insomnia | 119 (26.6) | 107 (29.2) | 12 (14.6) | .01 |
| Periodic limb movement disorder | 72 (16.1) | 61 (16.7) | 11 (13.4) | .47 |
| Restless legs syndrome | 87 (19.4) | 73 (19.9) | 14 (17.1) | .55 |
| Sleep-disordered breathing | 269 (60.0) | 217 (59.3) | 52 (63.4) | .49 |
| Other sleep disorders | 40 (8.9) | 29 (7.9) | 11 (13.4) | .11 |
| Neurologic disorders | ||||
| Convulsions | 45 (10.0) | 33 (9.0) | 12 (14.6) | .13 |
| Migraine | 79 (17.6) | 70 (19.1) | 9 (11.0) | .08 |
| Inflammatory disorders | ||||
| Psoriasis | 11 (2.5) | 8 (2.2) | 3 (3.7) | .43 |
| Inflammatory bowel disease | 13 (2.9) | 10 (2.7) | 3 (3.7) | .65 |
| Arthritis, inflammatory | 25 (5.6) | 23 (6.3) | 2 (2.4) | .17 |
| Gastrointestinal disorders | ||||
| Ulcer | 27 (6.0) | 26 (7.1) | 1 (1.2) | .04 |
| Gastritis | 72 (16.1) | 62 (16.9) | 10 (12.2) | .29 |
| Gastroesophageal reflux disease | 252 (56.2) | 202 (55.2) | 50 (61.0) | .34 |
| Skin | ||||
| Seborrhea | 18 (4.0) | 14 (3.8) | 4 (4.9) | .66 |
| Dermatophytosis | 19 (4.2) | 16 (4.4) | 3 (3.7) | .77 |
| Sensory/pain/other | ||||
| Dizziness and giddiness | 140 (31.2) | 106 (29.0) | 34 (41.5) | .03 |
| Neck pain | 133 (29.7) | 111 (30.3) | 22 (26.8) | .53 |
| Pain (other) | 25 (5.6) | 21 (5.7) | 4 (4.9) | .76 |
| Hearing loss | 53 (11.8) | 40 (10.9) | 13 (15.9) | .21 |
| Skin sensation abnormalities | 81 (18.1) | 72 (19.7) | 9 (11.0) | .06 |
Values are presented as n (%) yes. *Bold indicates significantly different prevalence (P < .05) among the groups. ICD = International Classification of Diseases, NOS = not otherwise specified, NPD = neurodegenerative parkinsonian disorders or dementia, RBD = rapid eye movement sleep behavior disorder.
Fourteen (17.1%) of those who were diagnosed with NPD met the 80% MDS prodromal threshold at the time of RBD code, when probability was calculated based on total LRPSG+. There was no difference in NPD diagnosis-free survival among those who met the 80% prodromal threshold at baseline vs those who did not, whether total LRPSG+ or total LRPSG− was included in the prodromal probability calculation (log-rank P = .49 or P = .657, respectively). We then examined other probability thresholds. Diagnosis-free survival among those who did vs did not meet the 50% prodromal threshold at baseline was not different based on the probability calculated with total LRPSG+. However, survival was significantly shorter among those who met the 50% threshold compared to those who were below the 50% threshold at baseline, using total LRPSG− (log-rank test P = .02).
Characteristics of the train and test samples are shown in Table 3. In the train dataset, in a Cox proportional hazard model that included all assessed individual MDS criteria risk factors and prodromal features (CoxMDS) present at baseline (time of first RBD code), greater hazards of NPD code was independently associated with male sex (hazard ratio = 2.06, 95% confidence interval = 1.04–1.10, P = .037), older age (hazard ratio = 1.07, 95% confidence interval = 1.05–1.10, P < .001), and cognitive dysfunction ICD code (hazard ratio = 2.83, 95% confidence interval = 1.79–4.46, P < .001) (Table 4). None of the other MDS criteria features, including RBD level of certainty, were independently significantly associated with risk of diagnosis. In the test dataset, time to NPD diagnosis among predicted high-risk individuals was significantly different from low-risk patients (Figure 1, 5-year survival free of NPD: 76% high risk vs 92% low risk, log-rank test P = .012). Specifically, median time to NPD diagnosis was shorter in the predicted high-risk vs low-risk group (7.67 vs 10.44 years, respectively). Harrell’s C for the CoxMDS model was 0.769, and Somers’ D was 0.538.
Table 3.
Characteristics of the train and test samples.
| Total | Training | Testing | P Value | |
|---|---|---|---|---|
| n = 448 | n = 359 | n = 89 | ||
| Diagnosed with NPD | 82 (18.3) | 66 (18.4) | 16 (18.0) | .93 |
| Age at first RBD code | 59.5 (15.4) | 59.9 (15.4) | 57.9 (15.6) | .27 |
| Individual MDS criteria features | ||||
| Male sex | 303 (67.6) | 244 (68.0) | 59 (66.3) | .76 |
| Possible iRBD | 258 (57.6) | 204 (56.8) | 54 (60.7) | .51 |
| Probable iRBD | 127 (28.3) | 110 (30.6) | 17 (19.1) | .03 |
| Definite iRBD | 63 (14.1) | 45 (12.5) | 18 (20.2) | .06 |
| Nicotine use | 204 (45.5) | 163 (45.4) | 41 (46.1) | .91 |
| Diabetes | 128 (28.6) | 105 (29.2) | 23 (25.8) | .52 |
| Gout, no or female | 422 (94.2) | 340 (94.7) | 82 (92.1) | .35 |
| Gout, yes in female | 26 (5.8) | 19 (5.3) | 7 (7.9) | |
| Anosmia | 12 (2.7) | 10 (2.8) | 2 (2.2) | .78 |
| Constipation | 97 (21.7) | 80 (22.3) | 17 (19.1) | .51 |
| Hypersomnolence | 43 (9.6) | 36 (10.0) | 7 (7.9) | .54 |
| Orthostasis | 35 (7.8) | 28 (7.8) | 7 (7.9) | .98 |
| Sexual dysfunction | 73 (16.3) | 55 (15.3) | 18 (20.2) | .26 |
| Urinary symptoms | 167 (37.3) | 134 (37.3) | 33 (37.1) | .97 |
| Depression | 199 (44.4) | 163 (45.4) | 36 (40.4) | .40 |
| Cognitive problems | 116 (25.9) | 95 (26.5) | 21 (23.6) | .58 |
| Putative risk factors/features | ||||
| Hypertension | 294 (65.6) | 234 (65.2) | 60 (67.4) | .69 |
| Dyslipidemia | 303 (67.6) | 244 (68.0) | 59 (66.3) | .76 |
| Head injury | 45 (10.0) | 40 (11.1) | 5 (5.6) | .12 |
| Alcohol disorder | 23 (5.1) | 18 (5.0) | 5 (5.6) | .82 |
| Tremor | 53 (11.8) | 47 (13.1) | 6 (6.7) | .10 |
| Gait abnormality | 102 (22.8) | 86 (24.0) | 16 (18.0) | .23 |
| Stiffness of joint | 13 (2.9) | 13 (3.6) | 0 (0.0) | .07 |
| Shoulder problem | 100 (22.3) | 87 (24.2) | 13 (14.6) | .05 |
| Osteoarthritis | 232 (51.8) | 188 (52.4) | 44 (49.4) | .62 |
| Anxiety/panic disorders | 220 (49.1) | 185 (51.5) | 35 (39.3) | .04 |
| Bipolar disorder | 32 (7.1) | 26 (7.2) | 6 (6.7) | .87 |
| Fatigue | 241 (53.8) | 198 (55.2) | 43 (48.3) | .25 |
| Sleep disturbance NOS | 81 (18.1) | 66 (18.4) | 15 (16.9) | .74 |
| Insomnia | 119 (26.6) | 102 (28.4) | 17 (19.1) | .08 |
| Periodic limb movement disorder | 72 (16.1) | 59 (16.4) | 13 (14.6) | .67 |
| Restless legs syndrome | 87 (19.4) | 72 (20.1) | 15 (16.9) | .49 |
| Sleep-disordered breathing | 269 (60.0) | 220 (61.3) | 49 (55.1) | .28 |
| Other sleep disorders | 40 (8.9) | 35 (9.7) | 5 (5.6) | .22 |
| Convulsions | 45 (10.0) | 32 (8.9) | 13 (14.6) | .11 |
| Migraine | 79 (17.6) | 65 (18.1) | 14 (15.7) | .60 |
| Psoriasis | 11 (2.5) | 8 (2.2) | 3 (3.4) | .53 |
| IBD | 13 (2.9) | 11 (3.1) | 2 (2.2) | .68 |
| Arthritis, inflammatory | 25 (5.6) | 20 (5.6) | 5 (5.6) | .99 |
| Ulcer | 27 (6.0) | 22 (6.1) | 5 (5.6) | .86 |
| Gastritis | 72 (16.1) | 61 (17.0) | 11 (12.4) | .29 |
| Gastroesophageal reflux disease | 252 (56.2) | 201 (56.0) | 51 (57.3) | .82 |
| Seborrhea | 18 (4.0) | 14 (3.9) | 4 (4.5) | .80 |
| Dermatophytosis | 19 (4.2) | 15 (4.2) | 4 (4.5) | .89 |
| Osteoarthritis | 232 (51.8) | 188 (52.4) | 44 (49.4) | .62 |
| Dizziness and giddiness | 140 (31.2) | 115 (32.0) | 25 (28.1) | .47 |
| Neck pain | 133 (29.7) | 106 (29.5) | 27 (30.3) | .88 |
| Pain | 25 (5.6) | 22 (6.1) | 3 (3.4) | .31 |
| Hearing loss | 53 (11.8) | 42 (11.7) | 11 (12.4) | .86 |
| Skin sensation abnormalities | 81 (18.1) | 72 (20.1) | 9 (10.1) | .03 |
Values presented as n (%) present or mean (standard deviation). IBD = inflammatory bowel disease, iRBD = isolated rapid eye movement sleep behavior disorder, MDS = Movement Disorders Society, NOS = not otherwise specified, NPD = neurodegenerative parkinsonian disorders or dementia, RBD = rapid eye movement sleep behavior disorder.
Table 4.
Results of Cox proportional hazards model when examining individual Movement Disorders Society criteria as predictors.
| Hazard Ratio | 95% CI Lower Bound | 95% CI Upper Bound | P Value | |
|---|---|---|---|---|
| Age at RBD ICD code | 1.067 | 1.039 | 1.096 | <.001 |
| Male sex | 2.061 | 1.045 | 4.062 | .037 |
| ICD code for: | ||||
| Nicotine abuse | 1.101 | 0.649 | 1.868 | .722 |
| Diabetes | 1.087 | 0.628 | 1.883 | .766 |
| Gout in male | 1.035 | 0.394 | 2.721 | .944 |
| Definite RBD | 0.819 | 0.398 | 1.683 | .586 |
| Olfactory dysfunction | 0.943 | 0.283 | 3.147 | .924 |
| Constipation | 1.490 | 0.804 | 2.760 | .205 |
| Hypersomnolence | 1.220 | 0.578 | 2.573 | .602 |
| Orthostasis | 0.486 | 0.177 | 1.333 | .161 |
| Hypersomnolence | 1.092 | 0.536 | 2.225 | .808 |
| Urinary dysfunction | 1.164 | 0.673 | 2.011 | .587 |
| Depression | 0.640 | 0.369 | 1.111 | .112 |
| Cognition | 2.966 | 1.780 | 4.941 | <.001 |
CI = confidence interval, ICD = International Classification of Diseases, RBD = rapid eye movement sleep behavior disorder.
Figure 1. Time to NPD diagnosis for predicted patients with high-risk vs low-risk iRBD.
ICD = International Classification of Diseases, iRBD = isolated rapid eye movement sleep behavior disorder, NPD = neurodegenerative parkinsonian disorders or dementia, RBD = rapid eye movement sleep behavior disorder.
In the Cox LASSO regression analysis, out of 49 features (Table 3), the model selected 8: age, sex, probable level of certainty for RBD, and ICD code for cognitive problems, insomnia, neck pain, skin sensation disturbance, and restless legs syndrome. In the test dataset, the time to NPD diagnosis among predicted high-risk patients was significantly different from low-risk patients (log-rank test P = .009). For the CoxLASSO model, Harrel’s C was 0.783 and Somers’ D was 0.566, indicating slight improvement in the predictive ability of the model.
Table 5 shows risk factors and prodromal features represented in EHR data besides ICD code. A family history of PD was documented in the family history tab data in only 9 patients. When data from the smoking history tab was incorporated with nicotine abuse ICD code, the proportion identified as former smoker only increased by 1% compared to ICD code alone. However, the smoking history tab data did identify 62 current smokers who would receive a greater protective LR, which the ICD code information cannot provide. Incorporating HbA1c did not identify additional patients with diabetes over those with diabetes ICD code alone, but did identify 14 individuals who received protective LR due to normal HbA1c. Finally, incorporating serum urate identified 11 additional individuals who received protective LR.
Table 5.
Movement Disorders Society criteria representations in EHR (beyond ICDs) and impact on criteria application for NPD diagnoses.
| Total | No NPD | NPD Diagnosed | P Value | ||
|---|---|---|---|---|---|
| n = 448 | n = 366 | n = 82 | |||
| First-degree family Hx | No data (LR = 1) | 439 (98.0%) | 361 (98.6%) | 78 (95.1%) | .040 |
| Yes (LR = 2.4) | 9 (2.0%) | 5 (1.4%) | 4 (4.9%) | ||
| Nicotine abuse ICD code only | No nicotine abuse (LR = 1) | 244 (54.5%) | 204 (55.7%) | 40 (48.8%) | .25 |
| Ever nicotine abuse (LR = 0.91) | 204 (45.5%) | 162 (44.3%) | 42 (51.2%) | ||
| Smoking history tab data | Never (LR = 1.2) | 88 (19.6%) | 73 (19.9%) | 15 (18.3%) | .065 |
| Former > 1 pack/y (LR = 0.91) | 182 (40.6%) | 138 (37.7%) | 44 (53.7%) | ||
| Current (LR = 0.51) | 62 (13.8%) | 56 (15.3%) | 6 (7.3%) | ||
| Former but < 1 pack/y (LR = 1) | 3 (0.7%) | 3 (0.8%) | 0 (0.0%) | ||
| No data (LR = 1) | 113 (25.2%) | 96 (26.2%) | 17 (20.7%) | ||
| Smoking by ICD code or smoking history tab data | Never (LR = 1.2) | 83 (18.5%) | 69 (18.9%) | 14 (17.1%) | .065 |
| Former > 1 pack/y (LR = 0.91) | 187 (41.7%) | 142 (38.8%) | 45 (54.9%) | ||
| Current (LR = 0.51) | 62 (13.8%) | 56 (15.3%) | 6 (7.3%) | ||
| Former but < 1 pack/y (LR = 1) | 3 (0.7%) | 3 (0.8%) | 0 (0.0%) | ||
| No data (LR = 1) | 113 (25.2%) | 96 (26.2%) | 17 (20.7%) | ||
| Diabetes ICD code only | No data (LR = 1) | 320 (71.4%) | 268 (73.2%) | 52 (63.4%) | .076 |
| Yes (LR = 1.5) | 128 (28.6%) | 98 (26.8%) | 30 (36.6%) | ||
| Diabetes by HbA1c only | No (LR = 0.95) | 14 (3.1%) | 13 (3.6%) | 1 (1.2%) | .55 |
| No data (LR = 1) | 354 (79.0%) | 288 (78.7%) | 66 (80.5%) | ||
| Yes (LR = 1.5) | 80 (17.9%) | 65 (17.8%) | 15 (18.3%) | ||
| Diabetes by ICD code or HbA1c | No (LR = 0.95) | 14 (3.1%) | 13 (3.6%) | 1 (1.2%) | .14 |
| No data (LR = 1) | 306 (68.3%) | 255 (69.7%) | 51 (62.2%) | ||
| Yes (LR = 1.5) | 128 (28.6%) | 98 (26.8%) | 30 (36.6%) | ||
| Serum urate | n/a (female, LR = 1) | 145 (32.4%) | 124 (33.9%) | 21 (25.6%) | .27 |
| Not measured in male (LR = 1) | 270 (60.3%) | 218 (59.6%) | 52 (63.4%) | ||
| Low in male (LR = 1.8) | 6 (1.3%) | 5 (1.4%) | 1 (1.2%) | ||
| High in male (0.88) | 27 (6.0%) | 19 (5.2%) | 8 (9.8%) | ||
| Gout ICD code only | No or female (LR = 1) | 422 (94.2%) | 346 (94.5%) | 76 (92.7%) | .52 |
| Yes in male (LR = 0.88) | 26 (5.8%) | 20 (5.5%) | 6 (7.3%) | ||
| Gout ICD code or serum urate | Yes in male (LR = 0.88) | 41 (9.2%) | 30 (8.2%) | 11 (13.4%) | .33 |
| No data or n/a (LR = 1) | 401 (89.5%) | 331 (90.4%) | 70 (85.4%) | ||
| No in male (LR = 1.8) | 6 (1.3%) | 5 (1.4%) | 1 (1.2%) | ||
| Estimated postprobability + EHR | 11.4 (2.2–41.0) | 8.4 (1.1–34.3) | 31.8 (14.2–63.7) | <.001 | |
| Prodromal probabilityICD+EHR >80% | No | 401 (89.5%) | 333 (91.0%) | 68 (82.9%) | .031 |
| Yes | 47 (10.5%) | 33 (9.0%) | 14 (17.1%) | ||
| Prodromal probabilityICD+EHR >50% | No | 359 (80.1%) | 301 (82.2%) | 58 (70.7%) | .018 |
| Yes | 89 (19.9%) | 65 (17.8%) | 24 (29.3%) | ||
| Prodromal probabilityICD+EHR >30% | No | 304 (67.9%) | 265 (72.4%) | 39 (47.6%) | <.001 |
| Yes | 144 (32.1%) | 101 (27.6%) | 43 (52.4%) | ||
EHR = electronic health record, HbA1c = glycated hemoglobin, Hx = history, ICD = International Classification of Diseases, LR = likelihood ratio, n/a = not applicable, NPD = neurodegenerative parkinsonian disorders or dementia.
The utility of ICD code for NPD diagnosis (ICD-defined phenoconversion) was contrasted to the method considered the gold standard for this analysis: clinically defined “phenoconversion” as ascertained via EHR manual chart review. The number of individuals with NPD diagnosis based on manual chart review who received a concordant ICD code was 23 (69.7%) patients with PD, 7 (58.33%) patients with dementia with Lewy bodies, 1 patient with multiple system atrophy, and 9 (28.12%) patients with other dementia. Among the 82 total patients determined to have NPD diagnosis by manual review, 45 (54.9%) received an ICD code for an NPD. Fourteen individuals diagnosed with NPD by chart review had more than 1 ICD code for an NPD. However, 7 patients who received the ICD code for NPD were determined to not have NPD on manual chart review. The sensitivity and specificity of ICD code for diagnosis based on manual chart review were 54.9% and 98.1%, respectively.
DISCUSSION
In this real-world sample of outpatients with iRBD, there was a high prevalence of prodromal nonmotor features. A diagnosis of NPD occurred in 82 (18%) patients. Using a model that included individual MDS criteria symptoms/signs and risk factors, we were able to discriminate those patients who were at high vs low risk of NPD. Incorporating other data from the EHR demonstrated the potential to leverage EHR data to further inform NPD risk in a patient with iRBD being seen in the clinic. However, our results indicate that to be useful, such models will need to incorporate EHR data beyond ICD codes that represent established or putative prodromal nonmotor and motor features and risk factors.
In this clinic-based sample of individuals with iRBD, we found the prevalence of meeting MDS criteria was much lower than in the research setting. In a study22 of 171 PSG-confirmed patients with iRBD, in which 7 risk markers and up to 8 prodromal features were assessed, 74% of patients with iRBD met the 2015 MDS criteria. In contrast, in our study, the prevalence of prodromal PD based on the 80% threshold was 10%; among the 82 individuals who developed NPD, 14 (17%) met MDS criteria. There was no difference in NPD diagnosis-free survival among those who met 80% probability for prodromal PD when PSG findings were factored into calculation of probability. These results reflect the challenges of applying MDS criteria to RBD participants in the real-world setting where prodromal features may or may not be systematically assessed and 40% of patients with iRBD did not have PSG confirmation.12,14 The MDS criteria are heavily weighted toward PSG confirmation of RBD; the LR for definite RBD is the highest of all MDS features (LR = 130). In contrast, presumed RBD but without PSG confirmation receives an LR of 2.8. Thus, prodromal probability in some individuals with iRBD may be underestimated. Indeed, as we previously reported14, incidences of NPD in patients with iRBD classified as probable (without PSG confirmation) or definite (with PSG confirmation) were similar, likely because many of the probable patients with iRBD were misclassified to a lower level of certainty. To further examine the influence of PSG confirmation on the MDS criteria in this real-world setting, we recalculated prodromal probability without accounting for PSG. Only 10 (2.2%) met the 80% threshold. Although there was no difference in NPD-free survival among those who met 80% probability when PSG was not factored into the calculation of probability, those meeting 50% probability for prodromal PD had shorter NPD-free survival. These findings indicate that it is possible to apply MDS criteria to identify relatively high-risk individuals with RBD even in the absence of PSG confirmation, but support the need for improved risk prediction in individuals with RBD without PSG confirmation.15
Leveraging EHR data to identify individuals at risk of NPD has the potential to facilitate translation of findings from research studies to clinical care. For example, this may help in designing clinical decision support tools that are built into the EHR to help inform primary care providers, who diagnose and manage about one-fourth of individuals with iRBD.12,14 Several studies on EHR data in various settings, including both in the United States and abroad across a variety of health care systems and EHR structures, support the utility of the EHR to identify prodromal features. In the United Kingdom, 2 case-control studies demonstrated consistent evidence of several prodromal features in individuals who went on to receive a PD diagnosis compared to those without.16,18 Five years prior to diagnosis, tremor, constipation, hypotension, erectile dysfunction, urinary dysfunction, and depression were different in cases vs controls (as were several other prodromal features that are not part of MDS criteria). These results were replicated in an outpatient primary care EHR dataset in Germany.23 Data from the US Veterans Affairs health care system also demonstrated similar findings in a different setting.24
Among individual MDS criteria components, age, male sex, and cognitive impairment were significantly independently associated with higher NPD risk in the multivariate Cox model. In part, these results may represent the large proportion of patients with dementia not otherwise specified among the NPD group. However, the findings may also add to the growing literature that cognitive changes are part of the PD prodrome: cognition has consistently been identified as a predictor of PD and other NPD in both prospective cohort studies25–27 and EHR data.16,18
Our findings indicate that in outpatients with iRBD, identifying prodromal features and risk markers using only ICD codes is feasible. Adding additional features, which can be computationally intensive and laborious to obtain, improved the model performance to some extent. This illustrates the potential for this approach to optimize prediction in iRBD if other features in the EHR could be added in the future. We also found that utilizing discrete data in the EHR to identify PD risk factors, namely, family history, smoking, physical activity, and caffeine, was of limited yield. Future work to determine prodromal features and risk factors through other components of the EHR, whether free text data from office visit notes, other laboratory tests, or physiologic assessments, will be critical. For example, a study of routinely obtained standard electrocardiograms in the 5 years prior to PD diagnosis identified a model from raw electrocardiographic data that had moderate accuracy in predicting PD.28
The lack of sensitivity of ICD codes for both prevalent and incident PD, NPD, and other neurodegenerative disorders has consistently been demonstrated. When comparing the ICD code in the EHR to Medicare claims data, specificity of ICD code for NPD was 99% and sensitivity was 61%.29 Few studies have examined ICD codes to identify phenoconversion. In our analysis, ICD code for NPD was not sensitive for predicting incident NPD diagnosis (“phenoconversion”), but was specific.
Limitations of this study include those inherent to conducting research with EHR data, including medical surveillance bias. EHR data may be biased toward individuals who are insured. Another challenge in data from the United States is that many individuals, whether those with private or public insurance, receive care across multiple health care systems.30,31 The availability of technology that allows linkage between EHRs across health care systems mitigates this to some extent, but siloed data remain substantial within EHRs in different systems. Our sample lacked racial and ethnic diversity. Thus, our results may not be generalizable. In addition, some prodromal features or risk factors may be present but not reflected in ICD codes or other discrete data available through the health records research query. In addition, we lacked a control group and cannot determine which prodromal features are specific to those with RBD.
We demonstrate a high prevalence of multiple prodromal nonmotor and motor features among outpatients with iRBD. The MDS criteria 80% threshold probability for prodromal PD has its limitations when PSG confirmation is lacking on a large proportion of individuals, as occurs in the real-world setting. However, a model that included established PD risk factors and prodromal features that contribute to the MDS criteria could accurately discriminate between individuals who were high vs low risk for diagnosis of NPD. Further, we show that ICD codes have some utility in identifying phenoconversion in the EHR, with excellent specificity but low sensitivity. Taken, together, our results demonstrate the potential of EHR data to translate research on prodromal PD to the clinic and may inform tools to aid clinicians in determining risk of NPD when they are evaluating a patient with iRBD. However, other components of EHR data beyond ICD codes will likely be needed in order to achieve this.
DISCLOSURE STATEMENT
All authors have seen and approved the manuscript. Work for this study was performed at the University of Pittsburgh. This work was funded by the University of Pittsburgh Dean’s Faculty Advancement Award (Chahine, principal investigator). University of Pittsburgh Medical Center Research Records Request (R3) is supported by the University of Pittsburgh’s Clinical and Translational Science Institute and the National Institutes of Health through Grant Number UL1 TR001857. The authors report no conflicts of interest.
ABBREVIATIONS
- CI
confidence interval
- CoxLASSO
Cox model that included only features selected by LASSO regularization as covariates
- CoxMDS
Cox model that included only individual components of the MDS criteria as covariates
- EHR
electronic health record
- HbA1c
glycated hemoglobin
- ICD
International Classification of Diseases
- iRBD
isolated rapid eye movement sleep behavior disorder
- LASSO
least absolute shrinkage and selection operator
- LR
likelihood ratio
- LRPSG+
likelihood ratio accounting for LR contributed by PSG confirmation of RBD
- LRPSG−
likelihood ratio not accounting for LR contributed by PSG confirmation of RBD
- MDS
Movement Disorders Society
- NPD
neurodegenerative parkinsonian disorders or dementia
- PD
Parkinson’s disease
- PSG
polysomnogram
- RBD
rapid eye movement sleep behavior disorder
- Total LRICD+EHR
total likelihood ratio incorporating information from ICD codes and other data from the EHR
REFERENCES
- 1. GBD 2021 Nervous System Disorders Collaborators . Global, regional, and national burden of disorders affecting the nervous system, 1990-2021: a systematic analysis for the Global Burden of Disease Study 2021 . Lancet Neurol. 2024. ; 23 ( 4 ): 344 – 381 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Chaudhuri KR , Azulay JP , Odin P , et al . Economic burden of Parkinson’s disease: a multinational, real-world, cost-of-illness study . Drugs Real World Outcomes. 2024. ; 11 ( 1 ): 1 – 11 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Cheng HC , Ulane CM , Burke RE . Clinical progression in Parkinson disease and the neurobiology of axons . Ann Neurol. 2010. ; 67 ( 6 ): 715 – 725 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Berg D , Crotty GF , Keavney JL , Schwarzschild MA , Simuni T , Tanner C . Path to Parkinson disease prevention: conclusion and outlook . Neurology. 2022. ; 99 ( 7 Suppl 1 ): 76 – 83 . [DOI] [PubMed] [Google Scholar]
- 5. Heinzel S , Berg D , Gasser T , et al. ; MDS Task Force on the Definition of Parkinson’s Disease . Update of the MDS research criteria for prodromal Parkinson’s disease . Mov Disord. 2019. ; 34 ( 10 ): 1464 – 1470 . [DOI] [PubMed] [Google Scholar]
- 6. Berg D , Postuma RB , Adler CH , et al . MDS research criteria for prodromal Parkinson’s disease . Mov Disord. 2015. ; 30 ( 12 ): 1600 – 1611 . [DOI] [PubMed] [Google Scholar]
- 7. Boeve BF , Silber MH , Ferman TJ , et al . Clinicopathologic correlations in 172 cases of rapid eye movement sleep behavior disorder with or without a coexisting neurologic disorder . Sleep Med. 2013. ; 14 ( 8 ): 754 – 762 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. American Academy of Sleep Medicine . International Classification of Sleep Disorders. 3rd ed . Darien, IL: : American Academy of Sleep Medicine; ; 2014. . [Google Scholar]
- 9. Postuma RB , Iranzo A , Hu M , et al . Risk and predictors of dementia and parkinsonism in idiopathic REM sleep behaviour disorder: a multicentre study . Brain. 2019. ; 142 ( 3 ): 744 – 759 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Iranzo A , Stefani A , Serradell M , et al. ; SINBAR (Sleep Innsbruck Barcelona) group . Characterization of patients with longstanding idiopathic REM sleep behavior disorder . Neurology. 2017. ; 89 ( 3 ): 242 – 248 . [DOI] [PubMed] [Google Scholar]
- 11. Postuma RB . Neuroprotective trials in REM sleep behavior disorder: the way forward becomes clearer . Neurology. 2022. ; 99 ( 7 Suppl 1 ): 19 – 25 . [DOI] [PubMed] [Google Scholar]
- 12. Havis I , Coates T , Wyant KJ , Spears CC , Garwood M , Kotagal V . Isolated REM sleep behavior disorder in North American older adults in an integrated health care system . J Clin Sleep Med. 2022. ; 18 ( 9 ): 2173 – 2178 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Bellomo G , De Luca CMG , Paoletti FP , Gaetani L , Moda F , Parnetti L . α-synuclein seed amplification assays for diagnosing synucleinopathies: the way forward . Neurology. 2022. ; 99 ( 5 ): 195 – 205 . [DOI] [PubMed] [Google Scholar]
- 14. Chahine LM , Ratner D , Palmquist A , et al . REM sleep behavior disorder diagnostic code accuracy and implications in the real-world setting . Neurol Clin Pract. 2025. ; 15 ( 1 ): e200387 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Mahlknecht P , Iranzo A , Stefani A , et al . Caveats of neurodegenerative risk stratification in idiopathic REM sleep behavior disorder by use of the MDS research for prodromal Parkinson’s disease . Sleep. 2017. ; 40 ( 10 ): zsx131 . [DOI] [PubMed] [Google Scholar]
- 16. Schrag A , Horsfall L , Walters K , Noyce A , Petersen I . Prediagnostic presentations of Parkinson’s disease in primary care: a case-control study . Lancet Neurol. 2015. ; 14 ( 1 ): 57 – 64 . [DOI] [PubMed] [Google Scholar]
- 17. Schrag A , Anastasiou Z , Ambler G , Noyce A , Walters K . Predicting diagnosis of Parkinson’s disease: a risk algorithm based on primary care presentations . Mov Disord. 2019. ; 34 ( 4 ): 480 – 486 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Simonet C , Bestwick J , Jitlal M , et al . Assessment of risk factors and early presentations of Parkinson disease in primary care in a diverse UK population . JAMA Neurol. 2022. ; 79 ( 4 ): 359 – 369 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Schrag A , Bohlken J , Dammertz L , et al . Widening the spectrum of risk factors, comorbidities, and prodromal features of Parkinson disease . JAMA Neurol. 2023. ; 80 ( 2 ): 161 – 171 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Camerucci E , Stang CD , Hajeb M , et al . Early-onset parkinsonism and early-onset Parkinson’s disease: a population-based study (2010-2015) . J Parkinsons Dis. 2021. ; 11 ( 3 ): 1197 – 1207 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Fereshtehnejad SM , Montplaisir JY , Pelletier A , Gagnon JF , Berg D , Postuma RB . Validation of the MDS research criteria for prodromal Parkinson’s disease: longitudinal assessment in a REM sleep behavior disorder (RBD) cohort . Mov Disord. 2017. ; 32 ( 6 ): 865 – 873 . [DOI] [PubMed] [Google Scholar]
- 22. Barber TR , Lawton M , Rolinski M , et al . Prodromal parkinsonism and neurodegenerative risk stratification in REM sleep behavior disorder . Sleep. 2017. ; 40 ( 8 ): zsx071 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Bohlken J , Schrag A , Riedel-Heller S , Kostev K . Identification of prodromal presentations of Parkinson’s disease among primary care outpatients in Germany . Neuroepidemiology. 2022. ; 56 ( 1 ): 41 – 49 . [DOI] [PubMed] [Google Scholar]
- 24. Scott GD , Lim MM , Drake MG , Woltjer R , Quinn JF . Onset of skin, gut, and genitourinary prodromal Parkinson’s disease: a study of 1.5 million veterans . Mov Disord. 2021. ; 36 ( 9 ): 2094 – 2103 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Speelberg DHB , Janssen Daalen JM , Bloem BR , Gagnon JF , Post B , Darweesh SKL . Prodromal cognitive deficits and the risk of subsequent Parkinson’s disease . Brain Sci. 2022. ; 12 ( 2 ): 199 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Chahine LM , Weintraub D , Hawkins KA , et al. ; PARS Investigators . Cognition in individuals at risk for Parkinson’s: Parkinson associated risk syndrome (PARS) study findings . Mov Disord. 2016. ; 31 ( 1 ): 86 – 94 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Weintraub D , Chahine LM , Hawkins KA , et al. ; PARS Investigators . Cognition and the course of prodromal Parkinson’s disease . Mov Disord. 2017. ; 32 ( 11 ): 1640 – 1645 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Karabayir I , Gunturkun F , Butler L , et al . Externally validated deep learning model to identify prodromal Parkinson’s disease from electrocardiogram . Sci Rep. 2023. ; 13 ( 1 ): 12290 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Searles Nielsen S , Warden MN , Camacho-Soto A , Willis AW , Wright BA , Racette BA . A predictive model to identify Parkinson disease from administrative claims data . Neurology. 2017. ; 89 ( 14 ): 1448 – 1456 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Williams BA . Constructing epidemiologic cohorts from electronic health record data . Int J Environ Res Public Health. 2021. ; 18 ( 24 ): 13193 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Raman SR , Curtis LH , Temple R , et al . Leveraging electronic health records for clinical research . Am Heart J. 2018. ; 202 : 13 – 19 . [DOI] [PubMed] [Google Scholar]

