Traceable Outcome Measures for Prodromal PD- Imaging view point

Ideas and Inspiration
1

Identifying a clinically meaningful and sensitive imaging marker for prodromal Parkinson’s disease (PD) is challenging because this stage lacks functional impairment that can serve as a clear clinical endpoint. As a result, outcome measures may serve different purposes. A measure that performs well for one purpose may be less useful for the others.

UPSIT illustrates this distinction. It is inexpensive, standardized, self-administered, and scalable, and hyposmia is an established prodromal risk marker, making it useful for cohort enrichment. However, olfactory impairment may already be substantial when individuals are identified, leaving limited room for measurable decline. In particular, participants with severe hyposmia or anosmia may show little longitudinal change because of floor effects. Olfactory scores are also influenced by age, smoking, sinonasal disease, and prior viral illness, including COVID-19. Thus, UPSIT may indicate who is at increased risk more effectively than how quickly an individual is progressing. Importantly, this limitation may depend on baseline severity: olfaction can continue to decline in individuals without anosmia, whereas scores may remain relatively stable among those who are already severely impaired.(Frontiers | Sequential change in olfaction and (non) motor symptoms: the difference between anosmia and non-anosmia in Parkinson’s disease).

MOCA has a different profile. It is brief, widely used, available in multiple languages, and familiar to clinicians and regulators. Early cognitive impairment, particularly in individuals with iRBD, can help identify those at increased risk of a more severe phenotype or dementia. However, many prodromal participants score within the normal range, limiting the sensitivity of the total score to subtle change. Practice effects, educational and cultural influences, and the coarse nature of a single global score further complicate interpretation. Domain-level ceiling effects have also been reported, particularly for naming and orientation, even when the total score shows little or no ceiling effect. MoCA may therefore be more useful for screening and stratification than as a stand-alone progression endpoint.pubmed.ncbi.nlm.nih

Other measures may address some of these limitations, including subtle motor signs assessed with the MDS-UPDRS Part III or timed tasks; digital and wearable measures of gait, tremor, sleep, and speech; dopamine-transporter imaging; autonomic measures; and fluid biomarkers such as neurofilament light chain (NfL) and α-synuclein seeding assays. Each has trade-offs. Digital measures can provide frequent, continuous assessments but remain heterogeneous in their analytic methods and clinical validation. Imaging is objective but expensive and not necessarily patient-relevant. NfL may reflect neurodegeneration but is not specific to synucleinopathy. α-synuclein seeding assays may offer greater biological specificity, although their role as longitudinal measures of progression remains under investigation.

Phenoconversion remains the most clinically meaningful endpoint, but it is event-driven and slow. In multicenter iRBD cohorts, the average annual risk of conversion to an overt neurodegenerative syndrome has been approximately 6%–7%, although rates vary with cohort composition and enrichment criteria. This creates substantial requirements for sample size, follow-up duration, and retention, especially for prevention trials.neurology+1

Several questions follow:

  • Could a composite measure combining several modestly informative outcomes outperform any single test?
  • Do different prodromal populations, such as iRBD, isolated hyposmia, LRRK2 or GBA1 carriers, and α-SAA-positive individuals, require different endpoint strategies?
  • How should sensitivity be balanced against participant burden over years of follow-up?
  • What should constitute a minimal clinically important difference in a population that may not yet have symptoms?
  • Should progression endpoints prioritize biological change, functional change, or the ability to predict phenoconversion?

A useful endpoint may therefore be context-dependent rather than universal: one measure for enrichment, another for monitoring progression, and a third for demonstrating clinical benefit. The central challenge is to determine which combinations of biomarkers are sufficiently sensitive, interpretable, and clinically meaningful for each prodromal population and study purpose.

2

Really interesting perspective! Thanks for laying out each of these markers so clearly, and their various pros and cons.

I’m all about data integration, so my first instinct is to agree with your first question about a composite measure. However, the method of creating a composite measure is very important! As an extreme example, if it were a simple sum then very few patients would reach a notable threshold, because each prodromal sub-population is likely experiencing a different mechanism that triggers only one of these markers. A naive approach would look for “3 out of 5 markers” or similar, but one sub-population patient could be fully anosmic while not triggering the other markers, and still be at higher risk.

If there truly is little overlap between the mechanisms of each prodromal sub-population, then the “5 weakly predictive markers” is likely to be just as good as combining them into a composite. But this is perhaps where there is still a research gap: how distinct ARE the mechanisms? And as you cover in your other questions, what should we be measuring as endpoints on each of these markers, in order to make meaningful predictions for patients?

All very interesting questions. I’d be very keen to hear what others think!

3

@vcatterson Thanks for this, and I agree that how the composite is built matters a great deal. A simple count of markers or a “3 out of 5” rule assumes the markers add up in one shared pathway, and your anosmic patient shows why that can fail.

A recent paper speaks to this (Kachhadia et al., npj Parkinson’s Disease, 2026, an All of Us matched case-control study of 2,673 PD cases and 10,692 controls). Two of its findings bear on your points:

1. The markers differ a lot in strength, so equal weighting would be a poor composite. REM sleep behavior disorder had an odds ratio of about 9.7 and mild cognitive impairment about 4.1. Orthostatic hypotension came in at about 1.8, while constipation, urinary dysfunction, depression and anxiety were all between 1.2 and 1.3. Insomnia showed no association. A weighted or tiered approach would likely beat a plain tally, and a high-specificity marker like RBD arguably shouldn’t need to be “added up” with anything.

2. Combined discrimination was modest. Using all nine non-motor features together, the model’s AUC was only 0.585. That suggests that in coded EHR data, merging weak markers doesn’t rescue individual-level prediction. It supports your point that you may do about as well with several weak markers as with a composite, at least when they’re limited to coded diagnoses. The authors conclude these features are informative at the group level but limited for individual prediction.

On your question about how distinct the mechanisms are, the paper’s exploratory K-means clustering of the features among cases is relevant. It didn’t find clearly separated phenotypes (silhouette widths of 0.43 to 0.53, unstable across k), and the authors say the subgroups shouldn’t be read as discrete prodromal phenotypes. The clusters that emerged were mostly low-burden, depression/anxiety-led, or urinary-led, and there was a small multi-feature group. So the data don’t support, or rule out, cleanly separate mechanisms. That is the research gap you describe, and EHR codes alone probably can’t close it.

A caveat that matters for your anosmia example: hyposmia wasn’t analysed in that study because it’s rarely coded. That is a good illustration of your point, since some of the most informative markers are never captured by routine records. The authors themselves argue that robust phenotypes will likely need olfactory testing, quantitative motor measures, biomarkers and genetic data (e.g. GBA, LRRK2), which fits your question about what endpoints we should be measuring for each marker.

4

Thanks for your detailed response, @AmgadDroby! That paper is extremely interesting, and I hadn’t seen it yet, so many thanks for sharing it. I agree with your interpretation and conclusions from the paper, especially that a weighted approach is likely to be most fruitful to risk prediction.

As I was reading, I was struck by the thought that many groups (including mine) are considering whether PD might be treated more effectively as a collection of separate diseases/mechanisms that have similar or overlapping symptom profiles. The type of results in this paper, and the direction of work you are suggesting moves even earlier in the timeline, considering the different mechanisms that give rise to the prodromal symptoms. I wonder if we will ultimately be able to use prodromal symptoms to predict future disease course and symptom constellation, once the interaction of different mechanisms is better understood.

I think that you have identified a key gap, which is the lack of recording of the less typical symptoms such as hyposmia in the first place. We can’t track, model, and predict what is not being recorded, and I suspect there may be a lot of under-counting of prodromal symptoms, especially if no drugs are prescribed. This is one reason why public outreach is so important, so that people know to raise symptoms with their family doctor, and doctors are aware of the breadth of symptoms that may be Parkinson’s-related.