Predicting Parkinson’s Dementia: Unlocking the Risk Factors

Hey there! Let’s talk about something pretty important and, frankly, a bit challenging: Parkinson’s disease. You know, it’s the fastest-growing neurological condition out there, second only to Alzheimer’s in terms of how common it is. While we often hear about the movement issues – the tremors, the stiffness – there’s another side that affects a lot of people: cognitive changes, and sadly, for many, dementia.

It turns out that about a quarter of people with Parkinson’s experience some mild cognitive impairment, and nearly half develop dementia within a decade of their diagnosis. That’s a significant number, and it really impacts quality of life, not just for the person with Parkinson’s, but for their loved ones and caregivers too. The big question is: why do some people develop dementia while others don’t? And can we see it coming?

Why Predicting PDD is a Big Deal

Predicting Parkinson’s Disease Dementia (PDD) isn’t just an academic exercise; it’s about real lives. PDD dramatically increases healthcare costs and, as I mentioned, severely reduces quality of life. It’s even considered one of the major milestones that often signals the later stages of the disease. Understanding who is at higher risk could potentially help healthcare teams prepare, support families, and maybe even explore ways to slow things down.

What Factors Might Be Involved?

So, what throws someone onto the path towards PDD? We know Parkinson’s itself is complex. While the cause is often a mystery, genetics definitely play a role – mutations in genes like LRRK2 and SNCA are linked to the disease. Environmental factors, like exposure to pesticides, can also increase risk, especially if there’s an underlying genetic susceptibility. And then there are lifestyle choices (diet, exercise, alcohol) and other health conditions, called comorbidities, like diabetes and high blood pressure, which have also been flagged as potential risk factors.

Previous studies have tried to connect specific genetic variants (like those in APOE, MAPT, GBA) to cognitive issues in Parkinson’s. Others have used machine learning with detailed clinical data to predict outcomes. But here’s the catch: a lot of these studies use data from specific research cohorts, which might not truly represent everyone with Parkinson’s out in the real world. Plus, some of the data used (like DNA methylation or spinal fluid biomarkers) isn’t exactly collected during a routine doctor’s visit. We really needed to look at factors that are more commonly recorded – things like other health issues, lifestyle habits, and basic genetic information – to see if we could build a prediction model that’s a bit more practical.

Our Approach: Diving into the Data

That’s where our study comes in. We wanted to tackle two main things using data from the UK Biobank (UKB), a massive health resource: first, could we predict dementia in people with Parkinson’s using machine learning? And second, could we figure out if any of the factors we looked at – especially the ones people might be able to change or manage – actually *cause* an increased risk of PDD? We also used data from the Parkinson’s Progression Markers Initiative (PPMI) study to see if our findings held up.

We used a few cool techniques. We trained machine learning models to predict PDD based on genetics, comorbidities, lifestyle, and environmental factors. To understand *why* the models made their predictions, we used Explainable AI (SHAP values). We also built Bayesian Networks to map out how all these different factors interact with each other. And finally, we employed Mendelian Randomization (MR) – a pretty neat genetic technique – to see if there were likely causal links between certain health conditions and PDD risk.

What Did We Find?

Okay, so what did the data tell us? We looked at thousands of people with Parkinson’s from the UKB. Our machine learning models (we tried a few, like Random Forest and XGBoost) had an average prediction performance of around 0.62 AUC. Now, I’ll be upfront, that’s not quite high enough for a doctor to use it alone in the clinic to say definitively “this person will get dementia,” but it could still be super useful for research, like helping to select participants for clinical trials.

The really interesting part came when we dug into *which* factors were most important for the predictions. Using SHAP analysis, we found that different types of data contributed, but genetics had the biggest influence overall (almost 50%!). Demographics (like age and sex) were next, followed by comorbidities.

Specific predictors that popped out included a polygenic risk score (a score that sums up the risk from many different genetic variants) for all-cause dementia, which was the single most important factor. Certain individual genetic variants (SNPs) were also significant, including one in the APOE gene locus, which is well-known for its link to Alzheimer’s. This is interesting because while Parkinson’s is characterized by alpha-synuclein protein clumps, many people with PDD also have Alzheimer’s-like pathology (amyloid-beta plaques and tau tangles), suggesting there might be different subtypes of PDD.

As you might expect, older age was a strong predictor – dementia is generally more common as people get older, and PDD typically develops years after the initial Parkinson’s diagnosis. But age alone wasn’t enough to predict PDD much better than chance, highlighting the need to consider multiple factors.

Other significant predictors included:

• Sex: Being male was associated with a higher predicted likelihood of PDD, consistent with other findings that men are more likely to develop it.
• BMI (Body Mass Index): Higher BMI actually seemed to be associated with a *lower* predicted likelihood of PDD. This might sound counterintuitive, but some studies suggest higher BMI could be protective against cognitive decline in Parkinson’s, although it could also be complicated by weight loss that happens in later stages of the disease.
• Comorbidities: Several health conditions were important predictors, including depression, anxiety, hypertension (high blood pressure), hypercholesterolemia (high cholesterol), and excessive daytime sleepiness. Depression and anxiety are common non-motor symptoms of Parkinson’s, and the others are known risk factors for dementia in general.
• Lifestyle: Even things like tea intake and water intake showed up as predictors. Previous research has hinted at green tea potentially being protective, while very high tea consumption might be linked to increased risk (maybe due to pesticides?). Lower water intake could be related to issues like difficulty swallowing (dysphagia) in people with Parkinson’s, and dehydration is known to affect cognition.

We also looked at the PPMI data, and while we had fewer variables available, the results were pretty similar, with age and the polygenic risk score still being key predictors.

Unpacking the Connections

Beyond just predicting, we wanted to see how these factors interact. Using Bayesian Networks, we mapped out the statistical dependencies. We saw expected links, like diabetes and obesity being connected to hypertension and high cholesterol. We also saw age linked to anxiety and daytime sleepiness. Interestingly, we found connections between genetic variants and non-genetic factors, like a specific genetic variant in the SNCA gene locus being linked to BMI, and other variants linked to air pollution exposure. This suggests a complex interplay between our genes and our environment/health status.

Pinpointing Causal Links

This is where the Mendelian Randomization (MR) came in. MR uses genetic variants as proxies to see if an exposure (like a comorbidity) likely *causes* an outcome (PDD), rather than just being associated with it. We looked at several comorbidities, but only hypertension and type 2 diabetes had enough suitable genetic proxies for the analysis.

Guess what? Our MR analysis indicated a *likely causal link* between both hypertension and type 2 diabetes and the risk of developing PDD. This wasn’t just an association; the genetic evidence suggests these conditions might directly contribute to PDD risk. This aligns with other studies showing that high blood pressure and poorly controlled blood sugar can negatively impact cognitive function and increase dementia risk, including in people with Parkinson’s.

What Does This Mean for Prevention?

The findings regarding hypertension and type 2 diabetes are particularly exciting because they are *modifiable* risk factors. While we can’t change our genes, we *can* manage our blood pressure and glucose levels. Our study suggests that actively managing these conditions in people with Parkinson’s might not just be good for their overall health, but could potentially help prevent or delay the onset of dementia. That’s a pretty powerful takeaway!

Acknowledging the Roadblocks

Now, as with any study, ours has limitations. We relied on diagnostic codes in the UK Biobank data, which aren’t always perfect and can sometimes mix up PDD with other types of dementia. We also used genotyping arrays, which capture common genetic variants but might miss rarer ones that could be important (like certain GBA variants). The UKB participants were also mostly white British, older, and predominantly male, so our findings might not apply perfectly to everyone with Parkinson’s globally. And while we used PPMI to validate some parts, we didn’t have enough data there to replicate the MR analysis on comorbidities.

Wrapping It Up

Despite the limitations, we think this study is a significant step. To our knowledge, we’re the first to really explore predicting PDD in a large population study like UKB using this combination of tools – machine learning, Bayesian Networks to see the connections, and Mendelian Randomization to look for causal links. While our prediction model isn’t ready for prime-time clinical use yet, it helped us confirm that genetics are the biggest player, but demographics and other health conditions (comorbidities) are also super important.

Crucially, our findings point to a likely causal link between hypertension and type 2 diabetes and PDD risk. This suggests that focusing on managing blood pressure and blood sugar in people with Parkinson’s could be a promising strategy for preventing or delaying dementia. There’s still lots more to learn, but understanding these risk factors is key to helping people live better with Parkinson’s.

Source: Springer