Polygenic risk scores and their limits

The research record

This is the project's file for this landmark, as written and checked. Sources are listed at the end; anything not yet confirmed is marked "Not yet verified".

Who

Valentina Escott-Price, Nigel Williams, John Hardy and the IPDGC (Annals of Neurology 2015); Mike Nalls, Cornelis Blauwendraat, Andrew Singleton and colleagues (270 authors, Lancet Neurology 2019; funded by the US National Institute on Aging, the Michael J. Fox Foundation and the Parkinson's Foundation); Kimberly Paul, Beate Ritz and colleagues (JAMA Neurology 2018).

What happened

A polygenic risk score (PRS) sums the effect of many common variants, each of which changes risk a little.

• 2015. Using 4,294 cases and 10,340 controls from a Parkinson's genome-wide meta-analysis (the work behind landmark 237), Escott-Price showed that a score from one study separated cases from controls in three independent studies. Patients with early onset had higher scores than those with late onset.
• 2018. In a population cohort of 285 patients, a 23-variant score was linked to faster cognitive decline (hazard ratio 1.44 per standard deviation) and faster motor decline (1.34 to 1.42). The lower ends of the confidence intervals were 1.00.
• 2019. The risk model in landmark 19 used 1,805 variants. Comparing the top with the bottom quarter, the odds ratio was 3.74 in the training sample (5,851 cases, 5,866 controls) and 6.25 in a validation sample of 527 cases and 472 controls. The area under the curve (AUC; 0.5 is chance, 1.0 is perfect) was 0.640 in training and 0.692 in validation.

Why it moved us toward a cure

A PRS helps pick people at higher risk for prevention trials and balance genetic risk across trial arms. It entered the prodromal risk model in landmark 222.

Outcome

Open. The authors of the 2019 study said screening the whole population is not feasible with the score alone: at 0.5% prevalence there would be about 14 false positives for each true case. The validation sample was small. All work used people of European ancestry; they warn it may not transfer. Later multi-ancestry scans (landmark 124) show shared loci, but I did not check how well the score performs outside Europe.

Sources

• Escott-Price et al., Ann Neurol 2015;77:582–591, DOI 10.1002/ana.24335
• Paul et al., JAMA Neurol 2018;75:360–366, DOI 10.1001/jamaneurol.2017.4206
• Nalls et al., Lancet Neurol 2019, DOI 10.1016/S1474-4422(19)30320-5 (numbers from the author manuscript, PMC8422160)
• No correction or retraction found for the three papers (Crossref and Europe PMC, 20 Oct 2026). The 2015 and 2018 abstracts and the 2019 manuscript's score section were read.
• Not yet verified: score performance in non-European groups; any newer score (post-2019); the 2015 study's effect size; clinical use.