Early Recognition of Treatment‐Responsive Rapidly Progressive Dementia: The Modified STAM3mP Score Authors: R W van Steenhoven, N Satyadev, A E M Bastiaansen, Y D Piura, J Kerstens, Y S Crijnen, J Brenner, T Brand, T M Bienfait, S Veenbergen, J M de Vries, F J de Jong, E G P Dopper, P A E Sillevis Smitt, H Seelaar, G S Day, M J Titulaer Published: 2026 ABSTRACT ABSTRACTEarly identification of patients with treatment‐responsive rapidly progressive dementia (RPD) is important as early treatment improves outcomes. The STAM3P score identifies treatment‐responsive RPD using “high risk” presenting features. We optimized performance by adding a time component (i.e., dementia within 3 months) and validated the modified “STAM3mP score” in two independent cohorts (n = 302). A STAM3mP score ≥ 1 identified 96% of patients with treatment‐responsive RPD; ≥ 3 features were highly specific (98%). These data suggest that ≥ 1 STAM3mP features effectively identifies treatment‐responsive RPD and may be used to prioritize investigations, whereas ≥ 3 STAM3mP features are highly specific and may inform early treatment decisions.Keywords: diagnostic score, rapidly progressive dementia, treatment‐responsive dementia Introduction Rapidly progressive dementia (RPD), defined as the development of dementia within one year or incapacitation within 2‐years of the onset of cognitive symptoms [1, 2, 3], requires a unique diagnostic approach, as underlying causes may be treatment‐responsive [1, 2, 4, 5]. However, the multitude of diagnostic tests and time pressure poses a challenge to clinicians [1]. Addressing this challenge is necessary as early identification of treatment‐responsive causes of RPD allows for early treatment, which is associated with better outcomes [6]. The STAM3P score was designed to identify patients with high likelihood of a treatment‐responsive causes of RPD at the first presentation based on the presence of clinically accessible features including seizures, tumor, age < 50‐years, mania, movement disorders, MRI suggestive of autoimmune encephalitis (AE), and CSF pleocytosis (≥ 10 cells/mm3) [1]. To date, the performance of this tool has been evaluated in a single U.S.‐based cohort, raising questions concerning generalizability. We evaluated the performance of the STAM3P score in an independent RPD cohort from the Netherlands. We further evaluated the performance of a modified criteria incorporating a time‐based component (i.e., dementia within 3‐months of symptom onset; STAM3mP score) and compared performance of the modified STAM3mP score to the original STAM3P score in our Dutch cohort and the original U.S. cohort. Methods Study Design The Dutch cohort was comprised of patients aged ≥ 18 years who developed dementia within one year of symptom onset [7], and were enrolled in a prospective multicenter study from 2019 to 2024. Three neurologists reviewed information on patient characteristics, clinical presentation, and ancillary testing. The U.S. RPD cohort was comprised of patients aged ≥ 18 years with RPD—defined as the development of dementia within one year or incapacitation within 2‐years of the onset of cognitive symptoms. Patients were prospectively enrolled within a study of RPD at Washington University in St. Louis (St. Louis, MO, USA) and Mayo Clinic in Florida (Jacksonville, FL, USA) from 2016 to 2022 [1, 5]. In both cohorts, patients with obvious causes of RPD (e.g., stroke, traumatic brain injury, infection) were excluded [1]. Final diagnoses were assigned referencing established diagnostic criteria [7, 8, 9, 10, 11]; causes of RPD were defined as potentially treatment‐responsive or non‐responsive following established methods [1]. Patients or legally authorized representatives provided written informed consent. Study procedures and policies were approved by institutional review boards at local institutions (Washington University in St. Louis, Mayo Clinic, and Erasmus University Medical Center). Statistical Analysis R statistical software (v4.3.2; R Core Team 2023) was used for statistical analysis. Pearson Chi‐Squared test or Fisher–Freeman–Halton test were used for group comparisons; two‐sided p < 0.05 was considered significant. The STAM3P score at first presentation was calculated by assigning one point for each of seizures, disease‐associated tumor, age of onset ≤ 50 years, mania, MRI suggestive of AE, movement disorders, and CSF pleocytosis (≥ 10 cells/mm3). The performance of cutoff values ≥ 1, ≥ 2, and ≥ 3 were evaluated [1]. Referencing previous literature that identified faster time to dementia onset as a characteristic of treatment‐responsive RPD [1, 3, 12], dementia within 3 months of symptom onset was evaluated as an additional feature associated with treatment‐responsive RPD in univariable and multivariable logistic regression analyses including STAM3P measures. This time‐based criterion was subsequently integrated into a modified STAM3mP score, which was evaluated within the Dutch and US RPD cohorts. The diagnostic performance of the original and modified scores was compared using area under curves (AUCs; DeLong test) and accuracy at established cutoffs (McNemar's test). Results Patient Characteristics The Dutch cohort included 147 patients; median age 67 years (range 36–86 years); 76 (46%) female. Ninety‐five of 147 (65%) patients were diagnosed with a potentially treatment‐responsive cause of RPD (Table 1). The frequency of neurodegenerative diseases was lower in the Dutch (14%) versus US cohort (22%; p = 0.041); other diagnoses were represented in similar proportions. TABLE 1Frequency of etiological diagnoses in patients with rapidly progressive dementia across Dutch and U.S. cohort.Etiological diagnostic categoriesTotal (n = 302)Dutch cohort (n = 147)US cohort (n = 155) P Treatment‐responsive 181 (60)95 (65)86 (55)0.079Autoimmune encephalitis109 (36)58 (39)51 (33)0.14Inflammatory CNS disorder28 (9)13 (9)15 (10)0.48Toxic/metabolic16 (5)9 (6)7 (5)0.36Neoplastic11 (4)5 (3)6 (4)0.54Primary psychiatric disorder8 (3)4 (3)4 (3)0.61Epilepsy (non‐inflammatory)4 (1)2 (1)2 (1)0.67Vascular2 (1)1 (1)1 (1)0.74CNS infection3 (1)3 (2)0 (0)0.11 Non‐responsive 121 (40)52 (35)69 (45)0.079Neurodegenerative diseases54 (18)20 (14)34 (22)0.041Creutzfeldt‐Jakob disease42 (14)17 (12)25 (16)0.16Vascular15 (5)8 (5)7 (5)0.46Neoplastic2 (1)1 (1)1 (1)0.74Other a 8 (3)6 (4)2 (1)0.13Abbreviation: CNS = central nervous system. a 6 unknown, 2 leukoencephalopathy. See Tables S3 and S4 for performance in individual cohorts. Frequency of etiological diagnoses in patients with rapidly progressive dementia across Dutch and U.S. cohort. Abbreviation: CNS = central nervous system. 6 unknown, 2 leukoencephalopathy. See Tables S3 and S4 for performance in individual cohorts. Modified STAM3mP Score In the combined cohort, dementia within 3 months of symptom onset was observed more frequently in treatment‐responsive diagnoses (56% vs. 31%, p < 0.001, Table S1) and was associated with treatment‐responsiveness in univariate (OR 2.93, 95% CI 1.80–4.76, p < 0.001) and multivariable logistic regression analyses (OR 3.09, 95% CI, 1.60–5.97, p = 0.001; Table S2). Accordingly, this feature was added to the original STAM3P score, creating the modified STAM3mP score. Modified STAM3mP Score: Dutch Cohort Overall performance, measured by AUC, between STAM3P and STAM3mP, increased from 0.79 to 0.83 (p = 0.058, Figure S1A). Accuracy, defined as the proportion of correctly classified cases with treatment‐responsive RPD, increased at the cutoff of ≥ 2 (63% vs. 76%; p = 0.016) and ≥ 3 (44% vs. 56%; p = 0.036; Table S3). Sensitivity increased significantly across all cutoff values: that is, STAM3mP improved detection of patients with potentially treatment‐responsive causes of RPD. The presence of ≥ 1 STAM3mP features identified 91/95 (96%) of patients with treatment‐responsive cases, with acceptable specificity (37%; Table S3). Modified STAM3mP ≥ 2 and ≥ 3 identified more patients with treatment‐responsive cases compared to STAM3P (73% vs. 47%; p = 0.001 and 34% vs. 14%; p < 0.001). Modified STAM3mP Score: US Cohort and Combined Cohort The AUCs of the original STAM3P and modified STAM3mP scores were similar in the U.S. cohort (both 0.88, p = 0.65; Figure S1B). However, sensitivity improved at cutoff values of ≥ 2 (62% vs. 83%; p < 0.001) and ≥ 3 (26% vs. 44%; p < 0.001), whereas accuracy increased at the cutoff value of ≥ 3 (59% vs. 68%; p < 0.001; Table S4). Increases in sensitivity came at the cost of decreased specificity, with specificity at cutoff values ≥ 1 decreasing from 62% (STAM3P) to 41% (modified STAM3mP; p < 0.001) and at cutoff values ≥ 2 from 93% to 83% (p < 0.001). When the modified STAM3mP score was applied to the combined cohort, overall AUC increased from 0.84 to 0.86 (p = 0.099, Figure S1C). Accuracy and sensitivity increased at cutoff values ≥ 2 and ≥ 3, at the cost of specificity, which decreased at cutoff values ≥ 1 and ≥ 2 (Table 2). TABLE 2Performance of the STAM3P and modified STAM3mP scores (Dutch and US RPD cohorts combined).STAM3PSTAM3mPTreatment‐responsiveNon‐responsiveAUC = 0.84Treatment‐responsiveNon‐responsiveAUC = 0.86 ≥ 1 16751218PPV 77% [73–80]17474248PPV 70% [67–73] < 1 147084NPV 83% [75–89]74754NPV 87% [76–93]181121302181121302Sens 92% [87–96]Spec 58% [49–67] Accuracy 78% [73–83] Sens 96% [92–98] Spec 39%* [30–48] Accuracy 73% [68–78] ≥ 2 989107PPV 92% [85–95]14020160PPV 88% [82–91] < 2 83112195NPV 57% [53–61]41101142NPV 71% [65–77]181121302181121302Sens 54% [47–62]Spec 93% [86–97] Accuracy 69% [64–75] Sens 77%* [71–83]Spec 83%* [76–90] Accuracy 80%*[75–84] ≥ 3 35035PPV 100% [90–100]70373PPV 96% [88–99] < 3 146121267NPV 45% [45–47]111118229NPV 52% [49–54]181121302181121302Sens 19% [14–26] Spec 100% [97–100] Accuracy 52% [46–57] Sens 39%* [32–46]Spec 98% [93–99] Accuracy 62%* [57–68] Abbreviations: NPV = negative predictive value; PPV = positive predictive value; Sens = sensivity. Spec = specificity.*Statistically significant difference (p < 0.05) between STAM3P vs. STAM3mP. Performance of the STAM3P and modified STAM3mP scores (Dutch and US RPD cohorts combined). Accuracy 78% [73–83] Sens 96% [92–98] Spec 39%* [30–48] Accuracy 73% [68–78] Accuracy 69% [64–75] Accuracy 80%*[75–84] Spec 100% [97–100] Accuracy 52% [46–57] Accuracy 62%* [57–68] Abbreviations: NPV = negative predictive value; PPV = positive predictive value; Sens = sensivity. Spec = specificity. Statistically significant difference (p < 0.05) between STAM3P vs. STAM3mP. The presence of ≥ 1 STAM3mP features showed highest sensitivity in patients with RPD attributed to AE (99%), toxic/metabolic causes (94%), and inflammatory central nervous system disorders (89%; Figure 1A, Table S5). Overall, 138/248 (56%) patients with a STAM3mP score ≥ 1 met possible AE criteria [9], of whom 100 (72%) were diagnosed with AE. Eight additional patients with AE were identified by an elevated STAM3mP score and subsequent autoantibody testing (see Table S6 for case examples demonstrating the clinical impact of the STAM3mP score). Performance of modified STAM3mP score in combined cohort (Dutch and the United States). A. Sensitivity of STAM3mP score in different potential treatment‐responsive diagnostic categories. *Other potential treatment‐responsive diagnostic categories: neoplastic, vascular, epilepsy, infectious, and psychiatric. B. Specificity of STAM3mP score in total cohort. Pie chart showing potential treatment‐responsive diagnostic categories in patients with ≥ 3 STAM3mP features. Of 73 patients with ≥ 3 STAM3mP features, 61 (84%) were diagnosed with AE or another inflammatory central nervous system disorder (Figure 1B). Three patients (4%) with CJD had three STAM3mP features, including dementia within 3 months (n = 3); movement disorders (n = 2); age at onset < 50 years (n = 1); a newly diagnosed tumor, initially considered disease associated (n = 1); seizures (n = 1); and CSF pleocytosis (10 cell/mm3, no erythrocytes; n = 1) that normalized on repeated lumbar puncture. Discussion Time from symptom onset to RPD ≤ 3 months was associated with treatment‐responsive causes of RPD. Accordingly, the addition of this feature to the existing STAM3P score improved detection of patients with treatment‐responsive RPD in independent cohorts, with only a modest increase in false positive detection: a modified STAM3mP score ≥ 1 recognized 96% of patients with treatment‐responsive RPD, with specificity of 39%. Given the potentially fatal consequences of missed diagnoses [13], we suggest that this sensitivity‐specificity trade‐off is acceptable and advocate that a STAM3mP score ≥ 1 may identify RPD patients who require further diagnostic evaluation for treatment‐responsive disorders [14]. Extended diagnostic work‐up should be individualized and may include neuronal autoantibodies testing in serum and CSF, additional neuro‐ and body imaging (e.g., PET), electroencephalography, and, in selected cases, brain biopsy [1, 15, 16]. The clinical AE criteria performed well in this study [9], demonstrating high sensitivity for AE (> 90%). These criteria complement the STAM3mP score and may be applied sequentially, with STAM3mP score used for initial screening for treatment‐responsive causes of RPD followed by AE criteria to inform next steps in the diagnostic evaluation and treatment of patients with possible AE. The presence of ≥ 2 and ≥ 3 STAM3mP features demonstrated high specificity (83% and 98%) for potentially treatment‐responsive forms of RPD in the combined cohort. Most patients with scores ≥ 3 (~85%) were diagnosed with autoimmune or inflammatory disorders—which typically warrant immunotherapy—suggesting that the modified STAM3mP scores ≥ 3 may be used to inform early treatment decisions (e.g., empiric immunotherapy). In RPD patients with ≥ 2 STAM3mP features, preliminary treatment may also be considered if an autoimmune/inflammatory cause of RPD is suspected and conditions where immunotherapy may reduce diagnostic yield or cause potentially harmful effects have been reasonably excluded (e.g., central nervous system infections, lymphoma, sarcoidosis) [1, 17, 18]. When considering preliminary treatment for a suspected etiologic diagnosis of AE, clinicians are cautioned to monitor for radiological “red flags” that may identify AE mimics, including persistent enhancement, restricted diffusion and mass effect on MRI [17, 19]. The STAM3mP score includes the presence of a disease‐associated tumor. Importantly, if initial tumor screening (e.g., CT scan) is negative, the detection of high‐risk tumor‐associated autoantibodies in blood or CSF still warrants comprehensive tumor screening/surveillance (e.g., repeated FDG‐PET) and, if necessary, oncological treatment [15, 20]. In summary, the presence of ≥ 1 STAM3mP features is useful to identify patients with RPD who are likely to benefit from further testing for treatment‐responsive causes, whereas scores ≥ 3 may identify patients who require further testing and may benefit from prompt initiation of immunomodulatory therapies. Author Contributions R.W.S., N.S., G.S.D., and M.J.T. contributed to the conception and design of the study; R.W.S., N.S., A.E.M.B., Y.D.P., J.K., Y.C., J.B., T.B., T.M.B., S.V., J.M.V., F.J.J., E.G.P.D., P.A.E.S.S., H.S., G.S.D., and M.J.T. contributed to acquisition and analysis of data; R.W.S., N.S., G.S.D., and M.J.T. contributed to drafting the text and preparing the figures. Funding The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Patient characteristics rapidly progressive dementia cohort treatment‐responsive vs. non treatment–responsive diagnoses (Dutch and U.S. cohort combined). Table S2: Univariate and multivariable logistic regression analysis of the STAM3mP score (Dutch and US RPD cohort combined). Table S3: Dutch RPD cohort: performance of the STAM3P and modified STAM3mP scores. Table S4: U.S. RPD cohort: performance of the STAM3P and modified STAM3mP scores. Table S5: Sensitivity of the modified STAM3mP score and possible AE criteria across diagnostic categories (U.S. and Dutch RPD Cohort combined). Table S6: Case examples of treatment‐responsive RPD demonstrating clinical application of the STAM3mP score. Figure S1: Comparison of Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) of STAM3P and modified STAM3mP across different rapidly progressive dementia cohorts (Dutch, the United States, and combined). Acknowledgments The authors thank all patients and their caregivers for their participation. P.A.E.S.S., J.M.V., M.J.T. of this publication are members of the European Reference Network for Rare Immunodeficiency, Autoinflammatory. and Autoimmune Diseases–Project ID No 739543 (ERN‐RITA; HCP Erasmus MC). Several authors (H.S., E.G.P.D. and F.J.J.) of this publication are members of the European Reference Network for Rare Neurological Diseases—Project ID No 739510. This study was funded by ZonMw (Memorabel Program) and Erasmus MC Foundation. RWvS has been funded by ZonMw (PARADE‐VIMP program), Alzheimer Nederland (InterAct Grant), and has received an Autoimmune Encephalitis Alliance Seed Grant (2024) and a Van Leersum Travel Grant from the Royal Dutch Academy of Science (2024). P.A.E.S.S. holds a patent for the detection of anti‐DNER, and received research support from Euroimmun; M.J.T. has filed a patent, on behalf of Erasmus MC, for methods for typing neurologic disorders and cancer, and devices for use therein, has received research funds for serving on a scientific advisory board of AmGen, for consultation at Guidepoint Global LLC, for consultation at UCB, has received an unrestricted research grant from Euroimmun AG and from CSL Behring, and has received funding from EpilepsieNL. This project was funded in part by the National Institute on Aging, National Institutes of Health K23AG064029 and R01AG089380 (GSD). van Steenhoven R. W., Satyadev N., Bastiaansen A. E. M., et al., “Early Recognition of Treatment‐Responsive Rapidly Progressive Dementia: The Modified STAM3mP Score,” Annals of Clinical and Translational Neurology 13, no. 9 (2026): 1956–1960, 10.1002/acn3.70434. R.W. van Steenhoven and N. Satyadev are shared first author. G.S. Day and M.J. Titulaer are shared senior author. Data Availability Statement Anonymized study data will be shared pending review of a request to the corresponding author from qualified individuals. References Associated Data Supplementary Materials Table S1: Patient characteristics rapidly progressive dementia cohort treatment‐responsive vs. non treatment–responsive diagnoses (Dutch and U.S. cohort combined). Table S2: Univariate and multivariable logistic regression analysis of the STAM3mP score (Dutch and US RPD cohort combined). Table S3: Dutch RPD cohort: performance of the STAM3P and modified STAM3mP scores. Table S4: U.S. RPD cohort: performance of the STAM3P and modified STAM3mP scores. Table S5: Sensitivity of the modified STAM3mP score and possible AE criteria across diagnostic categories (U.S. and Dutch RPD Cohort combined). Table S6: Case examples of treatment‐responsive RPD demonstrating clinical application of the STAM3mP score. Figure S1: Comparison of Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) of STAM3P and modified STAM3mP across different rapidly progressive dementia cohorts (Dutch, the United States, and combined). Data Availability Statement Anonymized study data will be shared pending review of a request to the corresponding author from qualified individuals.