The tool in question is the STAM3P score, first developed to identify patients whose rapidly progressive dementia is likely to have a treatment-responsive cause at their very first clinical presentation. The acronym captures seven clinically accessible features, each worth one point: seizures, a disease-associated tumor, age of onset of fifty years or younger, mania, magnetic resonance imaging findings suggestive of autoimmune encephalitis, movement disorders, and cerebrospinal fluid pleocytosis of at least ten cells per cubic millimeter. None of these features requires exotic technology; they can be gathered from the history, examination, standard MRI, and a routine lumbar puncture. Until now, however, the score’s performance had been evaluated in only a single United States-based cohort, leaving open the question of whether it would hold up in different patient populations and health care systems.

To answer that question, an international team led by researchers from Erasmus University Medical Center in the Netherlands, together with colleagues at Washington University in St. Louis and Mayo Clinic in Florida, tested the original score in an independent Dutch cohort and then asked whether it could be improved. Their reasoning was grounded in prior observations that treatment-responsive forms of rapidly progressive dementia tend to declare themselves faster than untreatable ones. Dementia that reaches diagnostic threshold within three months of symptom onset, they hypothesized, might itself be a marker of a reversible process. The Dutch cohort comprised 147 adults who developed dementia within one year of symptom onset and were enrolled prospectively across multiple centers between 2019 and 2024. The comparison group was the original American cohort of 155 patients enrolled from 2016 to 2022. Patients with obvious causes such as stroke, traumatic brain injury, or infection were excluded from both studies, and final diagnoses were assigned using established criteria by three neurologists reviewing each case.

The composition of the combined cohort of 302 patients underscores why this work matters. Sixty percent of all patients, 181 individuals, had a potentially treatment-responsive cause of their dementia. Autoimmune encephalitis alone accounted for 109 cases, or 36 percent, while inflammatory central nervous system disorders accounted for another 28. On the untreatable side, neurodegenerative diseases made up 18 percent and Creutzfeldt-Jakob disease 14 percent. In the Dutch cohort specifically, 65 percent of patients had a treatment-responsive diagnosis, a slightly higher proportion than the 55 percent seen in the American cohort, though the difference was not statistically significant. Neurodegenerative diseases were less frequent in the Dutch group, at 14 percent versus 22 percent, a difference the authors attribute in part to differing referral patterns and cohort definitions.

The statistical case for adding a time-based feature was strong. Dementia within three months of symptom onset occurred in 56 percent of patients with treatment-responsive diagnoses but only 31 percent of those with non-responsive causes, a highly significant difference. In multivariable logistic regression that accounted for the original STAM3P features, the three-month criterion carried an odds ratio of 3.09 for treatment-responsiveness, with a 95 percent confidence interval of 1.60 to 5.97. The investigators therefore created a modified score, dubbed STAM3mP, in which rapid progression to dementia within three months counts as an eighth point alongside the original seven features. They then compared the diagnostic performance of the two scores using area under the receiver operating characteristic curves, assessed with the DeLong test, and accuracy at established cutoffs, assessed with McNemar’s test.

The results in the Dutch validation cohort were encouraging. The area under the curve rose from 0.79 for the original score to 0.83 for the modified version, and accuracy improved significantly at cutoffs of two or more points, from 63 to 76 percent, and three or more points, from 44 to 56 percent. Most importantly, sensitivity increased across every cutoff value, meaning the modified score caught more of the patients whose dementia was actually treatable. A threshold of one or more STAM3mP features identified 96 percent of the 95 Dutch patients with treatment-responsive causes, at a specificity of 37 percent. In the American cohort, the overall discrimination was similar for both scores, at 0.88, but sensitivity again improved markedly at higher cutoffs, rising from 62 to 83 percent at the two-point threshold and from 26 to 44 percent at the three-point threshold. The price of this heightened sensitivity was a drop in specificity, from 62 to 41 percent at the one-point cutoff and from 93 to 83 percent at two points, a trade-off the authors argue is justified.

That argument rests on the consequences of a missed diagnosis. Autoimmune encephalitis and related inflammatory disorders can mimic prion disease and neurodegenerative dementia, and pathologically confirmed cases of autoimmune encephalitis in patients initially suspected of having Creutzfeldt-Jakob disease are well documented. When the modified score was applied to the combined cohort, the presence of at least one feature showed its highest sensitivity in patients with autoimmune encephalitis at 99 percent, toxic or metabolic causes at 94 percent, and inflammatory central nervous system disorders at 89 percent. Of the 248 patients with a score of one or more, 138 met criteria for possible autoimmune encephalitis, and 72 percent of those were ultimately confirmed to have the condition. Eight additional patients with autoimmune encephalitis were identified only after an elevated score prompted autoantibody testing, illustrating how the score can redirect the diagnostic pathway toward the right laboratory studies.

At the upper end of the scale, the modified score becomes not just a screening tool but a potential trigger for treatment. Among the 73 patients with three or more STAM3mP features, 84 percent were diagnosed with autoimmune encephalitis or another inflammatory central nervous system disorder, conditions that typically warrant immunotherapy. Only three patients with Creutzfeldt-Jakob disease reached that threshold, and even their features, such as mild pleocytosis that normalized on repeat lumbar puncture, illustrate the diagnostic noise the score must filter. With specificity of 98 percent at the three-point cutoff in the combined cohort, the authors suggest that such patients may be candidates for prompt initiation of empiric immunomodulatory therapy, provided that conditions in which immunotherapy could be harmful, such as central nervous system infection, lymphoma, or sarcoidosis, have been reasonably excluded. They also caution clinicians to watch for radiological red flags on MRI, including persistent enhancement, restricted diffusion, and mass effect, which can signal mimics of autoimmune encephalitis rather than the disease itself.

The study team proposes a practical, tiered algorithm. A score of one or more should prompt extended diagnostic workup, individualized to the patient, which may include neuronal autoantibody testing in serum and cerebrospinal fluid, additional neuroimaging and body imaging such as PET, electroencephalography, and in selected cases brain biopsy. A score of two or more raises the possibility of preliminary treatment when an autoimmune or inflammatory cause is suspected, while a score of three or more supports both further testing and early immunotherapy. The clinical criteria for autoimmune encephalitis performed well in this study, with sensitivity above 90 percent, and the authors recommend applying them sequentially after the score, using the STAM3mP as the initial filter and the criteria to guide the next steps. The score’s tumor component also carries a caveat: if initial screening is negative, detection of high-risk tumor-associated autoantibodies still warrants comprehensive and repeated tumor surveillance.

For a syndrome in which every week of delay narrows the window for recovery, a tool that costs nothing, requires no new tests, and can be computed at the bedside represents a meaningful advance. By validating the original STAM3P score in an independent European cohort and showing that a single, easily ascertained feature, the speed of progression to dementia, meaningfully improves its ability to detect treatable disease, the researchers have given neurologists a sharper instrument for one of medicine’s most time-sensitive diagnoses. The trade-off of more false positives, they contend, is one worth making when the alternative is a missed autoimmune encephalitis progressing irreversibly toward what looks like untreatable dementia.

Subject of Research: Validation and modification of the STAM3P clinical scoring tool for early recognition of treatment-responsive rapidly progressive dementia

Article Title: Early Recognition of Treatment‐Responsive Rapidly Progressive Dementia: The Modified STAM3mP Score

Article References: van Steenhoven, R. W., Satyadev, N., Bastiaansen, A. E. M., Piura, Y. D., Kerstens, J., Crijnen, Y. S., Brenner, J., Brand, T., Bienfait, T. M., Veenbergen, S., de Vries, J. M., de Jong, F. J., Dopper, E. G. P., Smitt, P. A. E. S., Seelaar, H., Day, G. S., & Titulaer, M. J. (2026). Early Recognition of Treatment‐Responsive Rapidly Progressive Dementia: The Modified STAM3 m P Score. Annals of Clinical and Translational Neurology, 13(9), 1956-1960. https://doi.org/10.1002/acn3.70434

Image Credits: AI Generated

DOI: 10.1002/acn3.70434

Keywords: rapidly progressive dementia, STAM3mP score, autoimmune encephalitis, diagnostic score, immunotherapy, Creutzfeldt-Jakob disease, cerebrospinal fluid, neurology, dementia, clinical validation, neuroinflammation, early diagnosis