Abstract
Global implementation of blood tests for Alzheimer’s disease (AD) would be facilitated by easily scalable, cost-effective and accurate tests. In the present study, we evaluated plasma phospho-tau217 (p-tau217) using predefined biomarker cutoffs. The study included 1,767 participants with cognitive symptoms from 4 independent secondary care cohorts in Malmö (Sweden, n = 337), Gothenburg (Sweden, n = 165), Barcelona (Spain, n = 487) and Brescia (Italy, n = 230), and a primary care cohort in Sweden (n = 548). Plasma p-tau217 was primarily measured using the fully automated, commercially available, Lumipulse immunoassay. The primary outcome was AD pathology defined as abnormal cerebrospinal fluid Aβ42:p-tau181. Plasma p-tau217 detected AD pathology with areas under the receiver operating characteristic curves of 0.93–0.96. In secondary care, the accuracies were 89–91%, the positive predictive values 89–95% and the negative predictive values 77–90%. In primary care, the accuracy was 85%, the positive predictive values 82% and the negative predictive values 88%. Accuracy was lower in participants aged ≥80 years (83%), but was unaffected by chronic kidney disease, diabetes, sex, APOE genotype or cognitive stage. Using a two-cutoff approach, accuracies increased to 92–94% in secondary and primary care, excluding 12–17% with intermediate results. Using the plasma p-tau217:Aβ42 ratio did not improve accuracy but reduced intermediate test results (≤10%). Compared with a high-performing mass-spectrometry-based assay for percentage p-tau217, accuracies were comparable in secondary care. However, percentage p-tau217 had higher accuracy in primary care and was unaffected by age. In conclusion, this fully automated p-tau217 test demonstrates high accuracy for identifying AD pathology. A two-cutoff approach might be necessary to optimize performance across diverse settings and subpopulations.
Similar content being viewed by others
Main
In recent years, notable progress has been made in identifying blood-based biomarkers (BBMs) for detecting Alzheimer’s disease (AD) pathology in symptomatic patients1. Among these BBMs, phosphorylated tau217 (p-tau217), measured alone or as a ratio relative to the concentration of nonphosphorylated tau217 (np-tau217), or Aβ42, has demonstrated superior performance across various methods, including immunoassays and mass spectrometry (MS)-based assays2,3,4,5,6,7. Despite these promising results, there are still a few challenges for the real-world implementation of BBMs into routine clinical practice. Some studies suggest that MS-based assays have high accuracy, comparable to US Food and Drug Administration (FDA)-approved cerebrospinal fluid (CSF) assays4,8, but their use is currently restricted to highly specialized laboratories. Conversely, fully automated immunoassay platforms offer a practical and scalable solution, leveraging the capacity of most clinical routine laboratories, which already rely on such platforms for diagnosing numerous disorders. However, there is a need to validate fully automated immunoassays for AD to make them more generally available in different clinical settings.
Diagnosis of AD is challenging, with up to 35% of patients potentially being misclassified in specialized clinics and 39% in primary care, when AD biomarkers are not used9,10. Most AD diagnoses are made by nonspecialist clinicians, such as primary care physicians, who are often the first to assess patients with cognitive symptoms but frequently have no access to currently available AD biomarkers, namely, CSF analysis or amyloid positron emission tomography (PET). The introduction of fully automated immunoassays may improve accessibility to these new biomarkers.
As anti-amyloid treatments for early symptomatic AD have become available in some countries, accurately identifying the disease’s etiology has become increasingly important because these treatments require biomarker confirmation of amyloid pathology. Furthermore, AD is often diagnosed in later stages, limiting opportunities for early intervention when the treatments are most effective. BBMs, with their broader accessibility, scalability and patient acceptability, hold the potential to provide more individuals with accurate and timely diagnoses, ensuring equitable access to appropriate care and new medications.
We have learned from CSF biomarkers that fully automated assays are essential for ensuring reproducibility of the results, particularly to facilitate the global adoption of these biomarkers11. A widely used CSF assay in clinical practice is the FDA-approved Lumipulse assay12. Recently, a plasma p-tau217 immunoassay was launched on the fully automated Lumipulse platform. Our aim was to validate this plasma p-tau217 assay for detecting AD pathology in clinical settings. Specifically, we established biomarker cutoffs in one cohort, applied them to patient samples that had been collected from independent primary and secondary care cohorts from different countries and examined the impact of comorbidities and demographic factors on the plasma p-tau217 performance. As secondary analyses, we compared the performance of plasma p-tau217 with that of the p-tau217:Aβ42 (Lumipulse) ratio currently under FDA evaluation13 and a high-performing MS-based method for plasma %p-tau217 (p-tau217:np-tau217 × 100) currently being used in clinical practice in the United States of America14.
Results
Participants and biomarker characteristics
Overall, 1,767 patients with cognitive symptoms were enrolled (n = 1,219 in secondary care; n = 548 in primary care). The mean age was 73 (s.d. 9) years, 53% were women and 55% had AD pathology (Table 1). Plasma p-tau217 concentrations were significantly higher in AD pathology-positive versus AD pathology-negative participants (Extended Data Fig. 1) and areas under the receiver operating characteristic (ROC) curves (AUCs) ranged from 0.93 to 0.96 across cohorts (Fig. 1d).
a–c,e–g, The single cutoff was set at >0.27 pg ml−1 (accuracy (a), PPV (b) and NPV (c)) and the two cutoffs at <0.22 and >0.34 pg ml−1 (accuracy (e), PPV (f) and NPV (g)). Comparisons between primary and pooled secondary care are shown in Supplementary Table 1. d, Note that the AUC values are independent of cutoffs. h, Participants who fell between the two cutoffs were classified as intermediate. Vertical dashed lines mark the performance in the Malmö cohort where the cutoffs were established. Data are presented as the observed percentage and the error bars as the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42:p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed (nmissing = 87 in primary care). The AD pathology prevalence was n = 153+ or 84− in Malmö, 93+ or 72− in Gothenburg, 321+ or 166− in Barcelona, 164+ or 66− in Brescia and 244+ or 305− in primary care (Sweden). Accuracy indicates percentage of correctly classified participants.
Accuracy of plasma p-tau217 (Lumipulse)
The rationale and methods for the cutoffs are provided in Methods. A single plasma cutoff was established in the Malmö cohort at 90% specificity (>0.27 pg ml−1), yielding an accuracy of 89% (95% confidence interval (CI), 86–93%) in the same cohort. Applying the cutoff to out-of-sample secondary care participants, the accuracy was 91% (95% CI = 86–95%) in Gothenburg, 89% (95% CI = 86–92%) in Barcelona and 89% (95% CI = 85–93%) in Brescia. In primary care (Sweden), the accuracy was 85% (95% CI = 82–88%) (Fig. 1a–c), which was significantly lower than in secondary care (Supplementary Table 1).
Next, we used a two-cutoff approach. The two cutoffs were established at 95% sensitivity and 95% specificity in the Malmö cohort (upper cutoff >0.34 pg ml−1, lower cutoff <0.22 pg ml−1). They yielded an accuracy of 93% (95% CI = 90–96%) in the same cohort with 17% in the intermediate zone (that is, between the upper and lower cutoffs and not included when calculating the accuracy). Applying these two cutoffs to out-of-sample cohorts resulted in an accuracy of 94% (95% CI = 90–98%) in Gothenburg, 94% (95% CI = 92–96%) in Barcelona and 92% (95% CI = 88–96%) in Brescia, with 12–14% in the intermediate zone (Fig. 1e–h). In primary care (Sweden), the accuracy was 92% (95% CI = 89–94%) with 16% in the intermediate zone. Using these two cutoffs, all positive predictive values (PPVs) and negative predictive values (NPVs) were ≥90%, except for the NPV in Brescia (84%, 95% CI = 75–92%), where it was reduced as a result of a high prevalence of AD pathology (71%; Table 1).
Effects of demographic factors and comorbidities
Among participants with AD pathology, plasma p-tau217 (Lumipulse) levels were significantly higher in women compared with men and higher in younger (aged <73 years) participants compared with older participants (aged 73–80 and ≥80 years) (Extended Data Fig. 2). Conversely, p-tau217 levels were lower in younger participants compared with older participants without AD pathology. In addition, p-tau217 concentrations were higher in participants with chronic kidney disease (CKD), irrespective of AD pathology status. There were no differences in p-tau217 concentrations when stratifying on APOE ε4 status (positivity defined as carrying at least one ε4 allele), educational level or presence of diabetes mellitus (Extended Data Fig. 2).
Despite the influence of some of these factors on p-tau217 concentrations, they did not significantly affect the accuracy of identifying AD pathology, except for age (Fig. 2a,c). Specifically, the accuracy was significantly lower in participants aged between 73 and 80 years (87%, 95% CI = 85–90%, P = 0.046) and aged ≥80 years (83%, 95% CI = 79–88%, P = 0.003) compared with those aged <73 years (91%, 95% CI = 88–92%). A numerically lower accuracy in older participants was observed across all cohorts, except for the Brescia cohort (Supplementary Table 2). The significant effect of age on accuracy persisted even after excluding participants with CKD, which is known to be more common in older individuals (Supplementary Table 2). When applying the two-cutoff approach, no significant effect of age or any other demographic factor or comorbidity was observed (Fig. 2b). However, in older individuals and those with CKD, the percentage of individuals with intermediate p-tau217 results increased to 19–20% (Fig. 2d). Examining the effects of older age and comorbidities in primary care, where they were more prevalent, revealed no significant effect on accuracy (Extended Data Fig. 3).
a–c, Accuracy using a single cutoff (a), two cutoffs (b) and AUC values (c). d, Participants with results between the two cutoffs classified as intermediate. The number of participants in each group, stratified by AD pathology, is indicated in a. Data are presented as the observed percentage and the error bars as the 95% CI derived from the bootstrap distribution. The analysis combined data from the five different cohorts (n = 1,767). The same analyses restricted to the primary care cohort can be found in Extended Data Fig. 3. AD pathology was defined as CSF Aβ42:p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed (n = 87). To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two sided) was calculated as the proportion of bootstrap resamples (n = 2,000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed using DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: 0.046, 0.003 (a); 0.035, 0.026 (c); 0.034, <0.001, <0.001 (d). aSignificantly higher than group 1, P < 0.05. bSignificantly higher than group 2, P < 0.05. cSignificantly higher than group 3, P < 0.05.
Accuracy across disease stages
When comparing Lumipulse p-tau217 accuracies and AUCs across groups with subjective cognitive decline (SCD), mild cognitive impairment (MCI) and dementia, no significant differences were found. Accuracies ranged between 86% and 89% using one cutoff and 92% and 94% using two cutoffs (Fig. 3a–c). As expected, based on the underlying prevalence of AD pathology, the NPV was higher in the SCD group whereas the PPV was higher in the dementia group. No significant differences in the percentage of intermediate results were found (Fig. 3d).
a–c, Using pooled data from all five cohorts (n = 1767), performance shown for participants with SCD (n = 250) (a), MCI (n = 858) (b) and dementia (n = 658) (c). Results are shown using a single cutoff (blue) or two cutoffs (red). Note that AUC values are independent of cutoffs. d, Participants who fell between the two cutoffs classified as intermediate. Data are presented as the observed percentage and the error bars as the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42:p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two sided) was calculated as the proportion of bootstrap resamples (n = 2,000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed using DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: 0.040, 0.001, 0.038 (a); <0.001, 0.046 (b); <0.001, 0.017, 0.024, 0.016 (c). aSignificantly higher than the SCD group, P < 0.05. bSignificantly higher than the MCI group, P < 0.05. cSignificantly higher than the dementia group, P < 0.05.
Comparison of plasma p-tau217 and the p-tau217:Aβ42 ratio
Plasma p-tau217:Aβ42 (Lumipulse) was available for 502 primary care participants and 911 secondary care participants (Supplementary Table 3). The p-tau217:Aβ42 levels were significantly higher in AD pathology-positive versus AD pathology-negative participants (Extended Data Fig. 4). The p-tau217:Aβ42 ratio provided similar accuracy and AUC compared with p-tau217 in primary and secondary care using a single cutoff (Fig. 4a–d). Using the two-cutoff approach, the accuracy of p-tau217:Aβ42 was significantly worse in primary care (Fig. 4e–g; P = 0.014). However, p-tau217:Aβ42 reduced the number of participants with intermediate results in both primary and secondary care from 15–16% to 7–10% (Fig. 4h; both P < 0.001). In addition, p-tau217:Aβ42 had a lower PPV and higher NPV in both primary and secondary care.
a–c,e–h, Pooled data from the secondary care cohorts (n = 911) and the primary care cohort (n = 502) examined. In the single cutoff approach (a–c). The cutoffs were >0.27 pg ml−1 for p-tau217 and >0.008 pg ml−1 for p-tau217:Aβ42. In the two-cutoff approach, the cutoffs for p-tau217 were <0.22 pg ml−1 and >0.34 pg ml−1 and <0.007 pg ml−1 and >0.009 pg ml−1 for p-tau217:Aβ42 (e–h). d, Note that the AUC values are independent of cutoffs. Cutoffs were established in the Malmö secondary care cohort (n = 337). Data are presented as the observed percentage and the error bars as the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42:p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two sided) was calculated as the proportion of bootstrap resamples (n = 2,000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed using DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: <0.001 (b); <0.001, 0.001 (c); 0.014 (e); <0.001, <0.001 (f); 0.007, 0.010 (g); <0.001, < 0.001 (h). *Significant difference between the two biomarkers (P < 0.05).
In contrast to plasma p-tau217, no increase in plasma p-tau217:Aβ42 ratio was observed in individuals with CKD regardless of AD pathology status or in older individuals without AD pathology (Extended Data Fig. 5). Similar to p-tau217, p-tau217:Aβ42 levels were increased in women and in younger participants with AD pathology status (Extended Data Fig. 5).
Comparison of Lumipulse and MS-based assays
MS-based p-tau217 and %p-tau217 data were available in a subset of patients from the Malmö (n = 337), Gothenburg (n = 164) and Brescia (n = 118) secondary care cohorts, as well as the primary care cohort (n = 513) (Supplementary Table 4 and Extended Data Fig. 6). No significant differences in AUCs or accuracies were found between the Lumipulse p-tau217 and MS-based p-tau217 or %p-tau217 assays in secondary care using either the single or the two-cutoff approaches (Fig. 5). In the primary care cohort, MS-based %p-tau217 had a significantly higher accuracy (90%, 95% CI = 87–92) and AUC (0.96, 95% CI = 0.94–0.97) than Lumipulse p-tau217 (accuracy = 85%, 95% CI = 82–88, P = 0.003; AUC = 0.92, 95% CI = 0.90–0.95, P = 0.006) and MS-based p-tau217 (accuracy = 85%, 95% CI = 82–88%, P < 0.001; AUC = 0.94, 95% CI = 0.92–0.96, P = 0.025) when using the single cutoff approach. The accuracy of Lumipulse p-tau217 in primary care was not statistically different from MS-based p-tau217 or %p-tau217 when using two cutoffs, but resulted in a significantly higher proportion of participants with intermediate results (Fig. 5h; all P < 0.003).
Data from the pooled secondary care cohorts (n = 619) and the primary care cohort (n = 513) were examined. The secondary care cohorts consisted of participants from the Malmö (n = 337), Gothenburg (n = 164) or Brescia (n = 118) cohort with MS-based data available. Cutoffs were set in the Malmö secondary care cohort (n = 337). a–c, In the single cutoff approach, the cutoffs were >0.27 pg ml−1 for Lumipulse p-tau217, >2.27 pg ml−1 for MS-based p-tau217 and >4.27 pg ml−1 for MS-based %p-tau217. e–h, In the two-cutoff approach, the cutoffs for Lumipulse p-tau217 were <0.22 pg ml−1 and >0.34 pg ml−1, <1.59 pg ml−1 and >2.92 pg ml−1 for MS-based p-tau217 and <3.55 pg ml−1 and >5.08 pg ml−1 for MS-based %p-tau217. Data are presented as the observed percentage and the error bars as the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42:p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two sided) was calculated as the proportion of bootstrap resamples (n = 2,000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed using DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: 0.003, <0.001 (a); <0.001 (b); 0.020, 0.004, <0.001 (c); 0.006, 0.025 (d); 0.020 (e); 0.008, 0.006 (f); 0.014 (g); 0.003, <0.001, <0.001, <0.001 (h). aSignificantly better than Lumipulse p-tau217, P < 0.05. bSignificantly better than MS-based p-tau217, P < 0.05.
Impact of patient characteristics and stage on assay accuracy
Among AD pathology-positive participants, %p-tau217 levels did not differ between any of the demographic or comorbidity groups. Specifically, the lower levels of Lumipulse p-tau217 and p-tau217:Aβ42 observed in the older age groups was not observed with MS-based p-tau217 and %p-tau217 (Extended Data Figs. 7 and 8). As age was the only demographic variable that showed differences in accuracy for Lumipulse p-tau217, we compared the accuracy and AUC of Lumipulse p-tau217 with MS-based p-tau217 and %p-tau217 across the three different age groups (Extended Data Fig. 9). As observed earlier, Lumipulse p-tau217 demonstrated lower accuracy in the older participants in this subsample (all P < 0.012), whereas no differences between age categories were observed for MS-based %p-tau217 (89% in all age groups).
There were no significant differences in accuracy, PPVs or NPVs between p-tau217 measured by Lumipulse and MS-based %p-tau217 across the three cognitive stages (SCD, MCI and dementia), regardless of whether one or two cutoffs were used (Extended Data Fig. 10). However, in the dementia group, %p-tau217 measured by MS demonstrated a significantly higher AUC compared with Lumipulse p-tau217 (P = 0.009). Furthermore, across all three cognitive groups, %p-tau217 classified significantly fewer individuals into the intermediate zone than Lumipulse p-tau217 (all P < 0.027).
Cost-effectiveness analysis of plasma p-tau217 (Lumipulse)
A cost-effectiveness analysis was performed comparing the Lumipulse p-tau217 assay with CSF analysis and amyloid PET imaging (Supplementary Table 5) and is available at https://bbrc-lab.shinyapps.io/Cost-effectiveness_analysis_plasma_p-Tau217. Using approximated costs for CSF AD biomarkers and amyloid PET imaging in the United States of America, we estimated savings of up to 60% compared with CSF testing alone and up to 81% compared with amyloid PET imaging alone, when implementing a two-cutoff approach with Lumipulse p-tau217. This calculation assumes that 14.7% of patients fall into the intermediate group (the total proportion in the present study) and require further testing (Supplementary Table 5). The app allows users to input region-specific costs for CSF and PET, enabling tailored cost analyses. A screenshot of the app is shown in Supplementary Fig. 1.
Sensitivity analyses
We performed additional analyses to assess the performance of Lumipulse p-tau217 and confirm the robustness of our results. First, the CSF Aβ42:40 ratio was used as a reference standard, instead of CSF Aβ42:p-tau181, with the FDA-approved cutoff of ≤0.072 pg ml−1 for positivity (Supplementary Fig. 2)12. Second, analyses were performed in the Swedish primary care cohort using only the CSF Aβ42:p-tau181 outcome, excluding those for whom Aβ pathology was defined by amyloid PET (n = 87; Supplementary Fig. 3). Third, a single cutoff established using the highest Youden index for AD pathology was used, instead of defining it at 90% specificity (Supplementary Fig. 4). The results of these sensitivity analyses were very similar to the main analysis.
Discussion
We found that the fully automated Lumipulse assay measuring plasma p-tau217 has a high accuracy in differentiating patients with AD pathology from those without. Several noteworthy aspects of the present study merit emphasis. First, the cutoffs were established in an independent cohort and then crossvalidated in several cohorts. Unlike many biomarker studies that develop ad-hoc plasma cutoffs specific to their cohort, our approach supports generalizability, facilitating broader use of the biomarker. This is of particular importance for tests intended for use in clinical practice using predefined cutoffs. Second, the high performance of the assay was consistent across four out-of-sample cohorts from different countries (Fig. 1), which included populations representing patients seeking advice for cognitive symptoms in specialized memory clinics and primary care. Moreover, the analyses of the plasma samples were performed in different centers and not centralized in one single laboratory. Third, we demonstrated that the accuracy was similar across different demographic, comorbidity and cognitive stage groups, except for a lower accuracy in older individuals, where a two-cutoff approach was necessary to achieve comparable accuracy (Figs. 2 and 3). Finally, we demonstrated, in a subset of patients, that the fully automated p-tau217 assay exhibits a performance comparable to MS-based measurements of p-tau217 in secondary care and slightly lower performance in primary care and older individuals, but not when using the two-cutoff approach (Fig. 5 and Extended Data Fig. 9). Given the suggested requirement of ≥90% accuracy for BBMs in the diagnosis of AD15, a two-cutoff approach for the Lumipulse p-tau217 assay might be necessary for its use as a stand-alone test for AD pathology, especially in primary care and across different age groups and cognitive stages (Figs. 1–3).
Although it has been known for several years that some BBMs can accurately detect AD pathology, these assays have not been widely available in most clinical centers and have been primarily limited to research cohorts. Therefore, the development of fully automated, user-friendly and reproducible tests is essential for this technology to become accessible to most centers. Lumipulse plasma assays have not previously been validated with predefined cutoffs and have been examined only in single-center studies16,17,18. A recent US-based study established a cutoff for p-tau217 at 0.25 pg ml−1 using the 90% specificity approach that we applied, which is comparable to the 0.27-pg ml−1 cutoff in our study17. For p-tau217:Aβ42, they only reported a cutoff at 85% specificity (0.008), which still aligns with our finding at 90% specificity (also 0.008). Similar to our study, they found that the p-tau217:Aβ42 ratio did not improve the accuracy but reduced the proportion with intermediate results. These results further support the generalizability of our cutoffs and results.
Comorbidities, especially CKD, are known to impact BBMs19,20. Notably, in our study the accuracy of Lumipulse p-tau217 was not affected by CKD. However, the presence of CKD resulted in a higher proportion of participants being classified into the intermediate group when two cutoffs were used (Fig. 2), presumably by the slight increase in p-tau217 concentration in participants without AD pathology (Extended Data Fig. 2d). This issue was mitigated by using the plasma p-tau217:Aβ42 ratio, with no increased levels observed among individuals with CKD (Extended Data Fig. 5d), consistent with previous findings for plasma ratios20. Another potential advantage of the plasma p-tau217:Aβ42 ratio was its higher NPV which could be especially important for primary care, although this comes at the cost of a lower PPV (Fig. 4). This phenomenon was observed in both primary and secondary care, and the only previous study reporting on this ratio observed a similar result using a Youden index cutoff for plasma p-tau217:Aβ42 (ref. 17).
The performance of Lumipulse p-tau217 significantly decreased with age (Fig. 2). This decline was not attributable to any specific cohort (Supplementary Table 2), including the primary care cohort (Extended Data Fig. 3), which had older participants and showed lower accuracy for Lumipulse p-tau217. The lower accuracy in the older individuals could potentially be explained by the lower p-tau217 concentration observed in the older participants with AD pathology, in combination with a higher p-tau217 concentration in older participants without AD pathology (Extended Data Figs. 2 and 5; lower levels in AD pathology-positive individuals were also observed for p-tau217:Aβ42). This could also explain the higher proportion of intermediate Lumipulse p-tau217 results in older participants when the two-cutoff approach was used (Fig. 2d). Previous studies have demonstrated that older individuals with AD presenting with cognitive symptoms in memory clinics often exhibit lower amounts of AD pathology, particularly tau pathology, partly as a result of the presence of other brain pathologies and lower cognitive reserve21. It is interesting that this age-related effect in AD pathology was not observed for %p-tau217 results, which showed consistent levels across age groups (Extended Data Fig. 8) and no age-related impact on accuracy (Extended Data Fig. 9). Other factors underlying this finding should be explored in future studies. Additional differences between Lumipulse p-tau217 and %p-tau217 included less influence of sex and CKD on %p-tau217, again highlighting the potential robustness provided by using a ratio (compare Extended Data Figs. 2, 7 and 8).
A key advantage of using BBMs is the potential to reduce the need for lumbar punctures, CSF analyses and amyloid PET scans, leading to substantial cost savings. This is especially timely because new disease-modifying drugs are becoming available in some countries, requiring biomarker confirmation of AD pathology. Given the considerable variation in costs for these tests across countries and centers, we created an online application to calculate the cost savings of using plasma p-tau217 (Lumipulse) compared with the local costs of CSF AD core biomarkers or amyloid PET (Supplementary Fig. 1). For example, using the approximate costs of CSF AD core biomarkers or amyloid PET in the United States of America, we estimated savings of up to 60% compared with solely using CSF tests and up to 81% compared with amyloid PET, using a two-cutoff approach and assuming that 14.7% of patients fall into the intermediate zone, which was the pooled proportion of intermediate results in the present study (Supplementary Table 5).
The present study has some limitations. First, our study included only European cohorts, necessitating further validation in clinical cohorts from other regions worldwide. Second, the slightly lower performance compared with %p-tau217 in primary care settings and among older age groups highlights the need for further investigation.
In summary, the utilization of a fully automated assay to measure plasma p-tau217, employing pre-established cutoffs, demonstrates high accuracy in detecting AD pathology across various specialized memory clinics and in primary care. These promising findings, coupled with the feasibility of implementing this technique in different settings, might facilitate the adoption of a BBM into routine clinical practice for more accurate AD diagnostics.
Methods
Participants
All participants were recruited as part of undergoing a memory investigation in clinical practice to ensure a representative, real-life study population. They all provided written informed consent. The studies were approved by the Swedish Ethical Review Authority (the Malmö, Gothenburg and primary care cohorts), the independent ethics committee, ‘Parc de Salut Mar’ Barcelona, Spain (the Barcelona cohort) and the Spedali Civili of Brescia local ethics committee, Italy (the Brescia cohort). The inclusion criteria for each cohort are described below.
In the BioFINDER-Memory Clinic study (NCT06122415)9, termed the Malmö cohort, the inclusion criteria were: (1) being under investigation for cognitive symptoms at the Memory Clinic of Skåne University Hospital, Sweden; and (2) CSF and blood sampling planned to be done as part of clinical practice even if the patient was not taking part in the present study. The exclusion criteria were: (1) not undergoing CSF or blood sampling as part of clinical practice and (2) not undergoing cognitive testing as part of clinical practice. Patients included in the present study were consecutively recruited between December 2022 and November 2023.
In the H70 Clinical Studies (termed the Gothenburg cohort), the inclusion criterion will be under investigation for cognitive symptoms at the Memory Clinic of Sahlgrenska University Hospital, Sweden. There were no exclusion criteria. For the present study, participants with CSF and blood sampling as part of the clinical investigation were included. Patients in the present study were consecutively recruited between March 2020 and June 2023.
In the BIODEGMAR study3 (termed the Barcelona cohort), the inclusion criteria were: (1) undergoing evaluation at the Cognitive and Behavioural Neurology Unit and inclusion in the DEGMAR register; (2) signed informed consent; and (3) having one of the following clinical diagnoses: SCD, MCI, AD dementia; behavioral variant frontotemporal dementia; progressive aphasia or primary progressive aphasia (logopenic, nonfluent and semantic variants); Lewy body dementia; corticobasal syndrome; progressive supranuclear palsy syndrome; and vascular cognitive impairment and dementia. Individuals with other causes of dementia, but unspecified clinical diagnoses, were also included and categorized as ‘other’. The exclusion criteria were: (1) age ≥80 years; (2) contraindication for lumbar puncture; or (3) disagreement with study procedures. Patients included in the present study were consecutively recruited between April 2017 and November 2023.
In the Life-BIO cohort (termed the Brescia cohort), the inclusion criteria were participants with MCI or mild dementia who underwent clinical routine CSF assessment at the outpatient neurodegenerative clinic of the Brescia University Hospital, Italy. The following exclusion criteria were applied: (1) cortical or subcortical cerebrovascular infarcts in structural imaging; (2) other neurological disorders or medical conditions potentially associated with cognitive deficits; (3) bipolar disorder, schizophrenia, history of drug or alcohol abuse or impulse control disorder; (4) recent traumatic events or acute fever or inflammation; and (5) refusal of collection of blood sampling for research purposes. Patients in this study were consecutively recruited between March 2020 and November 2023.
In the BioFINDER-Primary Care study (NCT06120361)9 (termed primary care cohort), the inclusion criteria were: (1) patient seeks medical help in primary care because of cognitive symptoms experienced by the patient or informant, or the primary care physician suspects a neurodegenerative disorder; (2) age ≥40 years; and (3) cognitive impairment characterized as SCD, MCI or mild dementia. The exclusion criteria were: (1) already diagnosed dementia; (2) substantial unstable systemic illness making it difficult to participate in the study; (3) current substantial alcohol or substance misuse; (4) refusing investigation at the memory clinic; (5) cognitive impairment with acute onset as a result of a stroke; and (6) cognitive impairment that can, with high certainty, be assessed by the primary care physician or explained by another condition or disease such as psychotic disorder, depression or alcohol abuse. Patients included in the present study were consecutively recruited from 19 primary care units in the south of Sweden from January 2020 to November 2023.
Criteria for SCD, MCI and dementia
In each cohort, syndromic diagnosis of SCD, MCI or dementia was done using established criteria22,23,24 or the Clinical Dementia Rating (CDR) scale25.
In the Malmö cohort9, SCD was defined as experiencing cognitive symptoms to the level that the patient was referred to the memory clinic owing to cognitive symptoms, but not fulfilling the criteria for MCI or dementia. MCI was defined as having cognitive symptoms and exhibiting objective cognitive impairment in any cognitive domain, based on the following neuropsychological battery: Trail Making Test A, Trail Making Test B and symbol digit modalities test (attention or executive function); verbal fluency animals, letter S fluency and the 15-word short version of the Boston naming test (verbal function); ten-word immediate and delayed recall from the AD assessment scale (ADAS), as well as a recognition task of the ten words (memory); and incomplete letters and cube analysis from the visual object and space perception battery (VOSP; visuospatial function). Objective cognitive impairment was not defined according to a strict threshold but was based on the overall clinical assessment, taking into account premorbid function. Dementia was diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edn (DSM-5) criteria for major neurocognitive disorder26.
In the Gothenburg cohort, SCD was defined as having cognitive symptoms and global CDR = 0, MCI as global CDR = 0.5 and dementia as global CDR ≥ 1 (ref. 25). CDR scoring was performed by experienced clinical staff at the memory clinic (physicians, nurses or neuropsychologists), based on routine examinations, patient interviews, reports from close relatives, cognitive tests (Mini-Mental State Examination (MMSE), Montréal Cognitive Assessment (MoCA) and the symbol digit modalities test), Lawton–Brody Instrumental Activities of Daily Living Scale (IADL)27, the Functional Activities Questionnaire28 and, in some cases, also neuropsychological testing (the Boston Naming Test, Processing Speed Identical Forms, Rey Auditory Verbal Learning Test (RAVLT), VOSP, Trail Making Test A and B, number repetition test, letter fluency, animal fluency test and Stroop and block design test).
In the Barcelona cohort, SCD was defined as subjective complaints of cognitive decline with normal performance on neuropsychological testing adjusted for age and educational level22. MCI was defined as clinical symptoms suggestive of cognitive decline with objective cognitive impairment in at least one cognitive domain on formal neuropsychological evaluation, with minor changes in ADLs, not meeting the criteria of dementia, and a global CDR score of 0.5. Dementia was defined as cognitive decline with objective impairment in at least one cognitive domain on neuropsychological evaluation and a CDR score of ≥1. The neuropsychological evaluation in the Barcelona cohort included the following cognitive tests and functional scales: MMSE, Memory Impairment Screen, Automatic reverse series (subtest of test Barcelona cognitive battery), semantic and phonetic fluency tasks (subtest of test Barcelona cognitive battery), Free and Cued Selective Reminding Test, Boston Naming Test, Trail Making Test, Blessed Dementia Rating Scale and Alzheimer’s Disease Functional Assessment and Change Scale.
In the Brescia cohort16, SCD was defined as self-perceived cognitive decline with normal performance on neuropsychological testing, adjusted for age and education and without functional impairment (CDR = 0). MCI was diagnosed based on objective cognitive deficits in at least one cognitive domain on extensive neuropsychological evaluation, with CDR = 0.5. Dementia was defined as a cognitive complaint with objective deficits in at least one cognitive domain on neuropsychological tests (CDR > 1). The neuropsychological battery assessment used in the Brescia cohort included the following cognitive tests: MoCA for global cognition, memory tests (Short story, Free and Cued Selective Reminding test or RAVLT), semantic and phonetic fluency tasks, naming test (SAND battery for aphasia), Trail Making Test, Digit Span forward and backward, Clock drawing test, Rey Complex figure copy and recall, and CDR.
In the primary care cohort9, SCD was defined as experiencing cognitive symptoms to the level that led the patient to seek help in primary care, but not fulfilling the criteria for MCI or dementia. MCI was diagnosed in weekly consensus rounds, including a responsible dementia specialist and neuropsychologist, based on the presence of notable cognitive symptoms and abnormal cognitive test results using the RBANS (Repeatable Battery for the Assessment of Neuropsychological Status) battery (accounting for premorbid cognitive level) and the CDR. The MCI definition did not require that a strict threshold in a cognitive domain was met (although all performed <−1 s.d. in at least one cognitive domain in the RBANS battery), but was based on the overall clinical assessment. The classification followed the design of the MCI classification of the Mayo Clinic Study of Aging29 and was in line with the DSM-5 criteria for mild neurocognitive disorder26. Dementia was diagnosed according to the DSM-5 criteria for major neurocognitive disorder26.
Plasma sampling and analysis
Plasma p-tau217 and Aβ42 were analyzed using the Lumipulse immunoassays (Fujirebio) at Gothenburg University, Sweden (the Malmö, Gothenburg and primary care cohorts), the Barcelonaβeta Brain Research Center, Spain (the Barcelona cohort) and the Department of Clinical Laboratory, ASST Spedali Civili Hospital, Italy (the Brescia cohort) in single batches. In the Malmö, Gothenburg, Brescia and primary care cohorts, p-tau217 and np-tau217 were also analyzed using MS-based assays at C2N Diagnostics, as previously described14. In addition to p-tau217, %p-tau217 (p-tau217:np-tau217 × 100) was also used. Additional details on plasma sampling and analyses are described in the Supplementary Information.
Comorbidities and demographic variables
Diabetes was defined as being diagnosed with either diabetes type 1 or diabetes type 2 based on registered diagnoses in the medical records. CKD was defined as an estimated glomerular filtration rate <60 ml−1 min−1 1.73 m−2. For age, participants were grouped as ≥80 years or <80 years. The <80-year group was further divided by a median split (73 years), yielding age categories <73 and 73–80.
Establishing plasma cutoffs
The cutoffs were established in the Malmö cohort according to a previously published design4. The cutoff was set at 90% specificity for AD pathology with as high a sensitivity as possible (one-cutoff approach). In addition, a two-cutoff approach (using upper and lower cutoffs) was also established according to the Alzheimer’s Association Appropriate use recommendations for AD blood biomarkers30 and previous publications1,9,31, and similar to the FDA approval of the CSF Aβ42:40 Lumipulse assays26. The two cutoff values were set to achieve 95% sensitivity (with maximized specificity) and 95% specificity (with maximized sensitivity), respectively, in the Malmö cohort4. Results between these two cutoffs were termed ‘intermediate’. To quantify the expected variability of cutoffs in external samples (that is, robustness of cutoffs), we used a bootstrap procedure (n = 2,000 iterations) to estimate the 95% CI of all cutoffs. The means of all bootstrap samples were used as the specific cutoff points. The cutoff for plasma p-tau217 (Lumipulse) at 90% specificity was >0.27 (95% CI = 0.20–0.31) pg ml−1. The two-cutoff approach at 95% sensitivity and 95% specificity yielded a lower cutoff at <0.22 (95% CI = 0.17–0.25) pg ml−1 and an upper cutoff at >0.34 (95% CI = 0.30–0.41) pg ml−1 for Lumipulse p-tau217. In the same cohort, this approach was also used for establishing cutoffs for p-tau217:Aβ42 (Lumipulse) and the MS-based methods. The cutoff for p-tau217:Aβ42 at 90% specificity was >0.008 (95% CI = 0.007–0.009). For the two-cutoff approach, the lower 95% sensitivity cutoff was <0.007 (95% CI = 0.006–0.008) and the upper 95% specificity cutoff was >0.009 (95% CI = 0.008–0.011). For the MS-based measures, the mean cutoff for p-tau217 at 90% specificity was >2.27 (95% CI = 1.83–2.54) pg ml−1. For the two-cutoff approach, the lower cutoff at 95% sensitivity was <1.59 (95% CI = 0.74–1.85) pg ml−1 and the upper cutoff at 95% specificity was >2.92 (95% CI = 2.37–3.59) pg ml−1. For %p-tau217, this approach yielded >4.27 (95% CI = 3.44–4.99) pg ml−1 at 90% specificity, <3.55 (95% CI = 3.06-4.37) pg ml−1 at 95% sensitivity and >5.08 (95% CI = 4.56-5.68) pg ml−1 at 95% specificity. In a sensitivity analysis, the single cutoff was established in the Malmö cohort at the highest Youden index for AD pathology.
Outcomes
The primary outcome was the presence of AD pathology, which was determined by the CSF Aβ42:p-tau181 ratio analyzed using the FDA-approved Lumipulse assays26. A cutoff of <11.94 for positivity was derived by unbiased Gaussian mixture modeling using all participants with CSF data (n = 1,679)32. For participants who could not undergo lumbar puncture in the BioFINDER-Primary Care study (n = 87), AD pathology was determined based on a positive visual read of [18F]flutemetamol PET. A sensitivity analysis was conducted, excluding individuals whose AD pathology was defined by amyloid PET. As a secondary outcome, the CSF Aβ42:40 ratio (Lumipulse) was used, with the FDA-approved cutoff of ≤0.072 indicating positivity26. Details of the CSF and PET analyses are described in the Supplementary Information.
Statistical analysis
There were no missing Lumipulse p-tau217 values or outcome data (AD pathology) for the included participants. As none of the statistical analyses depended on the assumption of a normal data distribution, raw biomarker levels were used throughout all analyses and are presented in the figures. Binary variables were compared using χ2 tests, and continuous variables using the Mann–Whitney U-test. ROC curves were used to calculate the AUCs. AUCs were compared using the DeLong test. Significant differences across accuracies, PPVs and NPVs were examined using a bootstrap hypothesis test (n = 2,000 bootstrap samples). The 95% CIs were calculated using bootstrapping (n = 2,000 resamples with replacement) and P values for significant differences in test metrics (for example, accuracy) were calculated as the proportion of bootstrap resamples (n = 2,000), where the absolute null-distributed statistic was greater than or equal to the observed difference. In the two-cutoff approach, participants with an intermediate test result were not considered when calculating the accuracy, PPVs and NPVs, but have instead been specified in the figures. A two-sided P < 0.05 was considered to indicate a statistical significance. R programming language (https://www.R-project.org, v.4.4.2) was used for all statistical analyses with the following packages: dplyr (v.1.1.4), pROC (v.1.18.5), boot (v.1.331), cutpointr (v.1.1.2), readxl (v.1.4.3), tidyverse (v.2.0.0), ggplot2 (v.3.5.1) and ggpubr (v.0.6.0).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Anonymized data will be shared by request from a qualified academic investigator for the sole purpose of replicating procedures and results presented in the article, as long as data transfer is in agreement with EU legislation on the general data protection regulation and decisions by the ethical review board of each site, which should be regulated in a material transfer agreement.
Code availability
The custom R codes for the analyses are available at GitHub (https://github.com/noellewarmenhoven/Lumipulse-Nat-Med).
References
Hansson, O., Blennow, K., Zetterberg, H. & Dage, J. Blood biomarkers for Alzheimer’s disease in clinical practice and trials. Nat. Aging 3, 506–519 (2023).
Ashton, N. J. et al. Diagnostic accuracy of a plasma phosphorylated tau 217 immunoassay for Alzheimer disease pathology. JAMA Neurol. 81, 255–263 (2024).
Ashton, N. J. et al. Plasma and CSF biomarkers in a memory clinic: head-to-head comparison of phosphorylated tau immunoassays. Alzheimers Dement. 19, 1913–1924 (2023).
Barthélemy, N. R. et al. Highly accurate blood test for Alzheimer’s disease is similar or superior to clinical cerebrospinal fluid tests. Nat. Med. 30, 1085–1095 (2024).
Janelidze, S. et al. Head-to-head comparison of 10 plasma phospho-tau assays in prodromal Alzheimer’s disease. Brain https://doi.org/10.1093/brain/awac333 (2022).
Palmqvist, S. et al. Discriminative accuracy of plasma phospho-tau217 for Alzheimer disease vs other neurodegenerative disorders. JAMA 324, 772–781 (2020).
Palmqvist, S. et al. Prediction of future Alzheimer’s disease dementia using plasma phospho-tau combined with other accessible measures. Nat. Med. 27, 1034–1042 (2021).
Warmenhoven, N. et al. A comprehensive head-to-head comparison of key plasma phosphorylated tau 217 biomarker tests. Brain 148, 416–431 (2025).
Palmqvist, S. et al. Blood biomarkers to detect Alzheimer disease in primary care and secondary care. JAMA https://doi.org/10.1001/jama.2024.13855 (2024).
Rabinovici, G. D. et al. Association of amyloid positron emission tomography with subsequent change in clinical management among Medicare beneficiaries with mild cognitive impairment or dementia. JAMA 321, 1286–1294 (2019).
Hansson, O. et al. CSF biomarkers of Alzheimer’s disease concord with amyloid-beta PET and predict clinical progression: a study of fully automated immunoassays in BioFINDER and ADNI cohorts. Alzheimers Dement. 14, 1470–1481 (2018).
EVALUATION OF AUTOMATIC CLASS Ill DESIGNATION FOR Lumipulse G 13-Amyloid Ratio (1-42/1-40) (Food and Drug Administration, 2022); www.accessdata.fda.gov/cdrh_docs/reviews/DEN200072.pdf
Fujirebio submits FDA regulatory filing for Lumipulse® G pTau 217/β-amyloid 1-42 plasma ratio in-vitro diagnostic test as an aid to identify patients with Amyloid pathology associated with Alzheimer’s disease. Fujirebio www.fujirebio.com/en/news-events/fujirebio-submits-fda-regulatory-filing-for-lumipulser-g-ptau-217bamyloid-142-plasma (2024).
Meyer, M. R. et al. Clinical validation of the PrecivityAD2™ blood test: a mass spectrometry-based test with algorithm combining %p-tau217 and Ab42/40 ratio to identify presence of brain amyloid. Alzheimers Dement. 20, 3179–3192 (2024).
Jack, C. R. Jr. et al. Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimers Dement. 20, 5143–5169 (2024).
Pilotto, A. et al. Plasma p-tau217 in Alzheimer’s disease: Lumipulse and ALZpath SIMOA head-to-head comparison. Brain 148, 408–415 (2025).
Figdore, D. J. et al. Optimizing cutpoints for clinical interpretation of brain amyloid status using plasma p-tau217 immunoassays. Alzheimers Dement. 20, 6506–6516 (2024).
Arranz, J. et al. Diagnostic performance of plasma pTau(217), pTau(181), Abeta(1-42) and Abeta(1-40) in the LUMIPULSE automated platform for the detection of Alzheimer disease. Alzheimers Res. Ther. 16, 139 (2024).
Mielke, M. M. et al. Performance of plasma phosphorylated tau 181 and 217 in the community. Nat. Med. 28, 1398–1405 (2022).
Janelidze, S., Barthelemy, N. R., He, Y., Bateman, R. J. & Hansson, O. Mitigating the associations of kidney dysfunction with blood biomarkers of Alzheimer disease by using phosphorylated tau to total tau ratios. JAMA Neurol. 80, 516–522 (2023).
Scholl, M. et al. Distinct 18F-AV-1451 tau PET retention patterns in early- and late-onset Alzheimer’s disease. Brain 140, 2286–2294 (2017).
Jessen, F. et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement. 10, 844–852 (2014).
McKhann, G. M. et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 7, 263–269 (2011).
Albert, M. S. et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging–Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 7, 270–279 (2011).
Morris, J. C. The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology 43, 2412–2414 (1993).
American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, 5th edn, 607–608 (APA, 2013).
Lawton, M. P. & Brody, E. M. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist 9, 179–186 (1969).
Pfeffer, R. I., Kurosaki, T. T., Harrah, C. H. Jr., Chance, J. M. & Filos, S. Measurement of functional activities in older adults in the community. J. Gerontol. 37, 323–329 (1982).
Roberts, R. O. et al. The Mayo Clinic Study of Aging: design and sampling, participation, baseline measures and sample characteristics. Neuroepidemiology 30, 58–69 (2008).
Hansson, O. et al. The Alzheimer’s Association appropriate use recommendations for blood biomarkers in Alzheimer’s disease. Alzheimers Dement. 18, 2669–2686 (2022).
Brum, W. S. et al. A two-step workflow based on plasma p-tau217 to screen for amyloid beta positivity with further confirmatory testing only in uncertain cases. Nat. Aging https://doi.org/10.1038/s43587-023-00471-5 (2023).
Benaglia, T., Chauveau, D., Hunter, D. R. & Young, D. S. mixtools: an R package for analyzing finite mixture models. J. Statist. Software 32, 1–29 (2009).
Acknowledgements
We would like to express our gratitude to the participants, and relatives, in all studies, without whom this research would have not been possible. In addition, we thank B. Arslan for analyzing plasma Aβ in the Gothenburg cohort. The BioFINDER group was supported by the National Institute of Aging (grant no. R01AG083740 to O.H., S.P., R.O. and S.J.), Alzheimer’s Association (grant nos. SG-23-1061717 and ZEN24-1069572 to O.H.), GHR Foundation (O.H. and S.P.), European Research Council (ERC, grant no. ADG-101096455 to O.H.), Swedish Research Council (grant nos. 2023-00356 to O.H., 2021-02219 to N.M.-C. and 2018-02052 to S.P.), ERA PerMed (grant no. ERAPERMED2021-184 to O.H.), the Knut and Alice Wallenberg Foundation (grant no. 2022-0231 to O.H.), the European Union’s (EU’s) Horizon Europe research and innovation program under grant agreement no. 101053962 (to O.H.), the Strategic Research Area MultiPark (Multidisciplinary Research in Parkinson’s disease) at Lund University (to O.H.), the Swedish Alzheimer Foundation (grant nos. AF-1011949 and AF-994075 to S.P., AF-994229 to N.M.-C. and AF-981132 to S.P.), the Swedish Brain Foundation (grant nos. FO2023-0163 to N.M.-C., FO2024-0284 and FO2022-0204 to S.P. and FO2021-0293 to O.H.), the family Rönström’s Foundation (grant nos. FRS-0003 to N.M.-C. and FRS-0004 and FRS-0011 to S.P.), the Berg Family Foundation (O.H.), the Parkinson Foundation of Sweden (grant no. 1412/22 to O.H.), the Cure Alzheimer’s fund (O.H.), the Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse (O.H.), the Skåne University Hospital Foundation (grant no. 2020-O000028 to S.P.), EU Joint Programme Neurodegenerative Diseases (grant no. 2019-03401 to N.M.-C.), WASP and DDLS Joint call for research projects (grant no. WASP/DDLS22-066 to N.M.-C.), Regionalt Forskningsstöd (grant no. 2022-1346 to S.P.) and the Swedish Federal Government under the ALF-agreement (grant nos. 2022-Projekt0107 to N.M.-C., 2022-Projekt0080 to O.H. and 2018-Projekt0226 to S.P.). The precursor of [18F]flutemetamol was sponsored by GE Healthcare. For the Gothenburg group, S.K. was financed by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement (nos. ALFGBG-1005471, ALFGBG-965923, ALFGBG-81392 and ALFGBG-771071), the Alzheimerfonden (grant nos. AF-842471, AF-737641, AF-929959 and AF-939825), the Swedish Research Council (grant nos. 2019-02075 and 2019-02075_15), Stiftelsen Psykiatriska Forskningsfonden and the Swedish Brain Foundation (grant no. FO2024-0097). A.D. was financed by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement (grant no. ALFGBG-984092). H.Z. is a Wallenberg Scholar and a distinguished professor at the Swedish Research Council. K.B. is supported by the Swedish Research Council (grant nos. 2017-00915 and 2022-00732), the Swedish Alzheimer Foundation (grant nos. AF-930351, AF-939721, AF-968270 and AF-994551), Hjärnfonden, Sweden (grant nos. ALZ2022-0006, FO2024-0048-TK-130 and FO2024-0048-HK-24), the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement (grant nos. ALFGBG-965240 and ALFGBG-1006418), the Alzheimer’s Association 2022–2025 (grant no. SG-23-1038904 QC), the Kirsten and Freddy Johansen Foundation, Copenhagen, Denmark, Familjen Rönströms Stiftelse, Stockholm, Sweden and an anonymous philanthropist and donor. The Barcelona group thanks P. Ortiz-Romero, M. de Diego, E. Jiménez, J. Torres and M. del Campo for technical support. F.A. received funding from grant no. JDC2022-049347-I, funded by the MCIU/AEI/10.13039/501100011033 and the EU NextGeneration EU/PRTR. M.S.-C. received funding from the ERC under the EU’s Horizon 2020 research and innovation program (grant no. 948677); ERA PerMed-ERA NET and the Generalitat de Catalunya (Departament de Salut) through project no. SLD077/21/000001; project PI19/00155 and PI22/00456, funded by Instituto de Salud Carlos III (ISCIII) and co-funded by the EU; and from a fellowship from ‘la Caixa’ Foundation (ID 100010434) and the EU’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie (grant no. 847648 (LCF/BQ/PR21/11840004)). For the Brescia group, A. Pilotto has beensupported by grants of Airalzh Foundation AGYR2021) Life-BIO grant, the LIMPE-DISMOV Foundation Segala Grant 2021, the Italian Ministry of University and Research PRIN COCOON (2017MYJ5TH) and PRIN 2021 RePlast (20202THZAW), the H2020 IMI IDEA-FAST (ID853981) and the Italian Ministry of Health, grant/award nos. RF-2018-12366209 and PNRR-Health PNRR-MAD-2022-12376110. V.Q. was supported by the H2020 IMI IDEA-FAST (grant no. ID853981). C.T. was supported by the PRIN 2021 RePlast (grant no. 20202THZAW). A. Padovani has been supported by grants of the Italian Ministry of University and Research PRIN COCOON (2017MYJ5TH) and PRIN 2021 RePlast (20202THZAW), the H2020 IMI IDEA-FAST (ID853981) and the Italian Ministry of Health, grant/award nos. RF-2018-12366209, RF-2019 RF-2019-12369272 and PNRR-Health PNRR-MAD-2022-12376110. The funders had no role in the design and conduct of the study, collection, management, analysis and interpretation of the data, preparation, review or approval of the paper and decision to submit the paper for publication.
Funding
Open access funding provided by Lund University.
Author information
Authors and Affiliations
Contributions
S.P., N.J.A., K.B., M.S.-C. and O.H. conceived and designed the project. S.P., N.W., F.A., M.S.-C., N.J.A. and O.H. drafted the paper. N.W. performed the statistical analyses. S.P., E.S., P.T., J.C., A.F.-L., A.P.-P., S.K., A.D., A. Pilotto and A. Padovani were responsible for clinical assessments. S.P., N.M.-C., R.O., H.Z., K.B., S.K., I.S., M.S.-C., A. Pilotto, A. Padovani and O.H. obtained funding. S.P., E.S., S.K., A. Pilotto, N.J.A., M.S.-C. and O.H. supervised the study. All authors helped out with acquisition, analysis and interpretation of data. All authors provided important intellectual content in the revision of the paper.
Corresponding authors
Ethics declarations
Competing interests
S.P. acquired research support (for the institution) from Avid and ADDF through ki elements. In the past 2 years, he received consultancy or speaker fees from BioArtic, Biogen, Eisai, Eli Lilly, Novo Nordisk and Roche. N.M.-C. received consultancy/speaker from Biogen, Owkin and Merck. R.S. received speaker fees from Roche and Triolab. H.Z. served on scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZpath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics and Wave, has given lectures in symposia sponsored by Alzecure, Biogen, Cellectricon, Fujirebio, Lilly, Novo Nordisk and Roche, and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside submitted work). R.O. received research funding or support from the ERC, ZonMw, NWO, National Institutes of Health, Alzheimer Association, Alzheimer Nederland, Stichting Dioraphte, Cure Alzheimer’s fund, Health Holland, ERA PerMed, Alzheimerfonden, Hjarnfonden, Avid Radiopharmaceuticals, Janssen Research & Development, Roche, Quanterix and Optina Diagnostics, has given lectures in symposia sponsored by GE Healthcare and is an advisory board member for Asceneuron and a steering committee member for Bristol Myers Squibb. All the aforementioned have been paid to the institutions. C.T. is supported by the Ministry of Health PRIN 2021 RePlast. A.P.‐P. served on advisory boards for Schwabe Farma Iberica. S.K. served on scientific advisory boards, and as a speaker and/or consultant for Roche, Eli Lilly, Geras Solutions, Optoceutics, Biogen, Eisai, Merry Life, Triolab, Novo Nordisk and BioArctic. A. Pilotto received travel grants from Abbvie, Bial, Lundbeck, Roche and Zambon pharmaceuticals, and personal compensation as a consultant or fees for lectures from Abbvie, Biogen and Lundbeck. A. Padovani received travel grants from Biougen, Lundbeck, Novonordisk, Roche pharmaceuticals, and personal compensation as a consultant or fees for lectures from Biogen, Lundbeck, Roche, Nutricia and General Healthcare (GE). K.B. served as a consultant and was on advisory boards for Acumen, ALZpath, AriBio, BioArctic, Biogen, Eisai, Lilly, Moleac Pte. Ltd, Novartis, Ono Pharma, Prothena, Roche Diagnostics and Siemens Healthineers, served at data monitoring committees for Julius Clinical and Novartis, has given lectures, produced educational materials and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai and Roche Diagnostics, and is a co-founder of BBS, which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. N.J.A. received consultancy or speaker fees from BioArtic, Biogen, Lilly, Quanterix and Alamar Biosciences. R.S. received a speaker’s fee from Roche. M.S.-C. received, in the past 36 months, consultancy or speaker’s fees (paid to the institution) from Almirall, Eli Lilly, Novo Nordisk and Roche Diagnostics. He received consultancy fees or served on advisory boards (paid to the institution) of Eli Lilly, Grifols, Novo Nordisk and Roche Diagnostics. He was granted a project and is a site investigator of a clinical trial (funded to the institution) by Roche Diagnostics. In-kind support for research (to the institution) was received from ADx Neurosciences, Alamar Biosciences, ALZpath, Avid Radiopharmaceuticals, Eli Lilly, Fujirebio, Janssen Research & Development, Meso Scale Discovery and Roche Diagnostics. M.S.-C. did not receive any personal compensation from these organizations or any other for-profit organization. O.H. is an employee of Eli Lilly and Lund University and previously acquired research support (for Lund University) from AVID Radiopharmaceuticals, Biogen, C2N Diagnostics, Eli Lilly, Eisai, Fujirebio, GE Healthcare and Roche. In the past 2 years, he received consultancy or speaker’s fees from ALZpath, BioArctic, Biogen, Bristol Meyer Squibb, Eisai, Eli Lilly, Fujirebio, Merck, Novartis, Novo Nordisk, Roche, Sanofi and Siemens. All C2N coauthors are salaried employees or consultants with cash and/or equity compensation from C2N Diagnostics. C2N Diagnostics performed the MS analyses blinded to any biomarker or clinical data and had no role in the statistical analysis or results. The other authors declare no competing interests.
Peer review
Peer review information
Nature Medicine thanks Aamna AlShehhi and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Jerome Staal, in collaboration with the Nature Medicine team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Boxplots of plasma p-tau217 (Lumipulse) concentrations in the different cohorts with AD pathology as grouping variable.
All p-tau217 levels differed significantly (P < 0.0001) between AD pathology-positive and AD pathology-negative participants in all cohorts, namely, Malmö (n = 337) (a), Gothenburg (n = 165) (b), Barcelona (n = 487) (c), Brescia (n = 230) (d), and in primary care (Sweden) (n = 548) (e). Horizontal dashed lines represent the single cutoff at 90% specificity established in (a) Secondary care Malmö. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD status was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. Abbreviations: AD, Alzheimer’s disease; IQR, interquartile range.
Extended Data Fig. 2 Plasma p-tau217 (Lumipulse) and the effect of demographic characteristics and comorbidities.
Boxplots of plasma p-tau217 (Lumipulse) concentrations stratified by AD status and the following variables: (a) sex, (b) APOE ε4 carriership, (c) education, (d), chronic kidney disease, (e) presence of diabetes, and (f) age. The analyses were performed pooling the data of the five cohorts (n = 1767). Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD status was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. Abbreviations: AD, Alzheimer’s disease; CKD, chronic kidney disease; IQR, interquartile range.
Extended Data Fig. 3 Effects of demographic characteristics and comorbidities on plasma p-tau217 performance in primary care.
This figure presents the accuracy and predictive values of plasma p-tau217 (Lumipulse) across various subgroups to assess the impact of demographic characteristics and comorbidities in primary care only (n = 548). Results are shown using a single cutoff (a) and using two cutoffs (c). Note that the AUC is independent of cutoffs (b). Participants who fell between the two cutoffs were classified in the intermediate group (d). Data are presented as the observed percentage, and the error bars as the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. The AD pathology prevalence is depicted for each cohort per subgroup in the figure (b). To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two-sided) was calculated as the proportion of bootstrap resamples (n = 2000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed with DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: 0.022; 0.002; 0.007. Abbreviations: Accuracy; percent correctly classified participants; CI, confidence interval; CKD, chronic kidney disease; NPV, negative predictive value; PPV, positive predictive value.
Extended Data Fig. 4 Boxplots of plasma p-tau217/Aβ42 (Lumipulse) concentrations with AD pathology as grouping variable.
Boxplots of plasma p-tau217/Aβ42 (Lumipulse) concentrations in the (a) pooled secondary care cohort (n = 911) and (b) primary care cohort (n = 502) with AD pathology as grouping variable. Plasma p-tau217/Aβ42 differed significantly ( P < 0.001) between AD positive and AD negative participants in both populations. Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. N = 10 outliers are not shown to improve visualization, but individuals were used in the analyses. One outlier (0.156) was AD pathology-negative. The remaining 9 outliers corresponded to AD pathology-positive participants, with the following plasma p-tau217/Aβ42 levels: 0.223, 0.256, 0.260, 0.273, 0.287, 0.509, 0.671, 0.761, 0.872. All outliers had MCI or dementia. Abbreviations: AD, Alzheimer’s disease; CKD, chronic kidney disease; IQR, interquartile range.
Extended Data Fig. 5 Plasma p-tau217/Aβ42 (Lumipulse) and the effect of demographic characteristics and comorbidities.
Boxplots of plasma p-tau217/Aβ42 concentrations stratified by AD pathology and the following variables: (a) sex, (b) APOE ε4 carriership, (c) education, (d), chronic kidney dysfunction, (e) presence of diabetes, and (f) age. The analyses were performed pooling the plasma p-tau217/Aβ42 data of the five cohorts (n = 1413). Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. N = 10 outliers are not shown to improve visualization, but individuals were used in the analyses. One outlier (0.156) was AD pathology-negative. The remaining 9 outliers corresponded to AD pathology-positive participants, with the following plasma p-tau217/Aβ42 levels: 0.223, 0.256, 0.260, 0.273, 0.287, 0.509, 0.671, 0.761, 0.872. All outliers had MCI or dementia. Abbreviations: AD, Alzheimer’s disease; CKD, chronic kidney disease; IQR, interquartile range.
Extended Data Fig. 6 Boxplots of mass spectrometry-based plasma p-tau217 and %p-tau217 with AD pathology as grouping variable.
Pooled secondary care cohort data (a, c, n = 619) and primary care data (b, d, n = 513) with AD pathology as grouping variable to assess differences in MS-based p-tau217 (a and b) and %p-tau217 (c and d). Plasma levels differed significantly (P < 0.0001) between AD pathology-positive and AD pathology-negative participants in both populations. Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. Abbreviations: AD, Alzheimer’s disease; MS, mass spectrometry; IQR, interquartile range.
Extended Data Fig. 7 Mass spectrometry-based plasma p-tau217 and the effect of demographic characteristics and comorbidities.
Concentrations stratified by AD pathology and the following variables: (a) sex, (b) APOE ε4 carriership, (c) education, (d), chronic kidney dysfunction, (e) presence of diabetes, and (f) age. The analyses were performed pooling the data of the four cohorts with MS data available (n = 1132). Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond to the first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. Abbreviations: AD, Alzheimer’s disease; CKD, chronic kidney disease; MS, mass spectrometry; IQR, interquartile range.
Extended Data Fig. 8 Mass spectrometry-based plasma %p-tau217 and the effect of demographic characteristics and comorbidities.
Plasma %p-tau217 stratified by AD pathology and the following variables: (a) sex, (b) APOE ε4 carriership, (c) education, (d), chronic kidney dysfunction, (e) presence of diabetes, and (f) age. The analyses were performed pooling the data of the four cohorts with MS data available (n = 1132). Dashed lines represent the single cutoff at 90% specificity established in the Malmö secondary care cohort. The dots represent individual participants. The central band of the boxplot represents the group median, the box limits correspond tothe first and third quartiles, and the whiskers represent the minimum/maximum value or the 1.5 interquartile range (IQR) from the box, whichever is smaller. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive amyloid PET visual read for those who did not undergo lumbar puncture. Group mean levels were compared with Student’s t-test. Abbreviations: AD, Alzheimer’s disease; CKD, chronic kidney disease; MS, mass spectrometry; IQR, interquartile range.
Extended Data Fig. 9 Comparison of assay performances between different age groups.
This figure presents the accuracy and predictive values of plasma p-tau217 (Lumipulse) and p-tau217 (MS) and %p-tau217 (MS) across three age groups to examine the impact of demographic characteristics and comorbidities. The analysis combined data from four different cohorts with MS data available (n = 1132). Results are shown using a single cutoff. The dots represent the actual percentage, and the error bars the 95% CI derived from the bootstrap distribution. AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The Pvalue (two-sided) was calculated as the proportion of bootstrap resamples (n = 2000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed with DeLong statistics. Results were not corrected for multiple comparisons. The comparisons between age groups per biomarker are shown in the figure (P values in the order as presented in the plot: 0.012; <0.001; 0.013). %p-tau217 (MS) had significantly higher AUCs than p-tau217 (MS) when comparing the age categories 73-80 (P = 0.033) and ≥80 (P = 0.013). No differences were observed between the biomarkers within the different age categories. Abbreviations: Accuracy, percent correctly classified participants; CI, confidence interval; MS: mass spectrometry; NPV, negative predictive value; PPV, positive predictive value.
Extended Data Fig. 10 Performances of plasma p-tau217 (Lumipulse) vs. mass spectrometry-based plasma p-tau217 and %p-tau217 across disease stages.
This figure presents the accuracy and predictive values of plasma p-tau217 (Lumipulse) and plasma p-tau217 (MS) and %p-tau217 (MS) across the disease stages SCD (a), MCI (b) and dementia (c). Results are shown using a single cutoff (blue, a-c) and two cutoffs (red, SCD (d), MCI (e) and dementia (f). Note that AUC values are independent of cutoffs (g). Participants who fall between the two cutoffs were classified in the intermediate group (h). The dots or bars represent the actual percentage, and the error bars the 95% CI. The analysis combined data from four different cohorts with MS data available (n = 1132). AD pathology was defined as CSF Aβ42/p-tau181 < 11.94 or positive visual read on amyloid PET if lumbar puncture was not performed. To assess whether the observed difference in the statistics is significantly different from zero, we performed a bootstrap hypothesis test. The P value (two-sided) was calculated as the proportion of bootstrap resamples (n = 2000) where the absolute null-distributed statistic was greater than or equal to the observed difference. Differences between AUCs were assessed with DeLong statistics. Results were not corrected for multiple comparisons. Significant P values in the order as presented in the plot: (b) 0.043; 0.046, (d) 0.033; 0.024, (g) 0.003; 0.009; 0.006, (h) 0.027; <0.001; <0.001; 0.005; <0.001. Abbreviations: CI, confidence interval; MS: mass spectrometry; MCI, mild cognitive impairment; NPV, negative predictive value; PPV, positive predictive value; SCD, subjective cognitive decline.
Supplementary information
Supplementary Information (download PDF )
Supplementary Methods, Tables 1–5 and Figs. 1–4.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Palmqvist, S., Warmenhoven, N., Anastasi, F. et al. Plasma phospho-tau217 for Alzheimer’s disease diagnosis in primary and secondary care using a fully automated platform. Nat Med 31, 2036–2043 (2025). https://doi.org/10.1038/s41591-025-03622-w
Received:
Accepted:
Published:
Version of record:
Issue date:
DOI: https://doi.org/10.1038/s41591-025-03622-w
This article is cited by
-
Clinical translation of fluid, imaging, and digital biomarkers for Alzheimer’s disease
Alzheimer's Research & Therapy (2026)
-
Shake and bake: a robust and cost-effective proteomic sample preparation workflow for plasma and cerebrospinal fluid
Clinical Proteomics (2026)
-
Plasma p-tau217, quantified by the fully automated LUMIPULSE G platform, outperforms p-tau181 in predicting amyloid pathology in cognitive complaints patients
Scientific Reports (2026)
-
Prevalence of Alzheimer’s disease pathology in the community
Nature (2026)
-
Advances in Alzheimer’s disease: mechanistic insights and therapeutic targets
Science China Life Sciences (2026)
