EEG is a method of recording brain activity using electrophysiological indicators, which is the basis for the generation of ERPs (Randolph and Helfrich, 2019; Donoghue and Voytek, 2022). In 1875, Richard Caton et al. recorded electrical activity on the surface of the exposed brain of rabbits, 1924 Hans Berger first recorded EEG on the scalp of patients with skull injuries and healthy people (Randolph and Helfrich, 2019; Berger, 1929). After the discovery of EEG, scientists tried to study evoked potentials (EP) using stimulation-induced brain waves. Sensory evoked potentials were first recorded on the scalp of the awake human brain by Pauline and Hallowell Davis in 1935–1936 (Davis et al., 1939). With the deepening of research, it was found that EP can be caused not only by external stimulation but also by active top-down psychological factors, so “stimulation” was changed to “event” and “evoked potential” was changed to “event-related potential” (Randolph and Helfrich, 2019). In 1962, Galambos and Sheatz published the first ERPs paper on computer-averaged superposition (Galambos and Sheatz, 1962); In 1964, Grey Walter et al. discovered the first cognitive ERPs component (contingent negative variation, CNV) (Walter et al., 1964), marking the beginning of a new era in ERPs research; In 1965, Sutton, Braren, Zubin, and John discovered the P300 (Sutton et al., 1965), more and more researchers began to explore the components of ERPs related to cognition.

The components of ERPs are related to the brain’s information processing of stimuli, which can be divided into exogenous components related to the processing of physical stimuli and endogenous components related to mental processing. The endogenous components are mismatch negativity (MMN), P300 (Sutton et al., 1965), CNV, and N400 (Kutas and Hillyard, 1980), which reflect different higher cognitive processing processes. In addition, the components of ERPs also include movement-related potentials (BSP, MP, RAF), lateralized readiness potential (LRP), processing negativity (PN), recognition potential (RP), error-related negativity (ERN, Pe, FN), visual C1, P1, visual N2, etc. (Donoghue and Voytek, 2022; Sutton et al., 1965; McWeeny and Norton, 2020; Pratt and Kappenman, 2011; Pratt, 2011; Friedman et al., 2001; Kappenman et al., 2021). Table 1 summarizes the various ERP components and associated properties.

ERPs reflect the integrated cognitive functions of the brain, including the processes of information extraction, processing, and output, etc. The greatest advantage of ERPs in neuroscience is its high temporal resolution, which can accurately and without delay respond to the brain’s activity at each moment in time, and they are widely used in the assessment of cognitive impairment in various central nervous system (CNS) diseases as well as in the diagnosis of psychiatric and psychological disorders.

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) flow diagram.¹

The PRISMA flow diagram is presented in Figure 1.

A literature search in PubMed, Web of Science and SCOPUS was performed in March 2024 using the following search string: PubMed: (“Alzheimer Disease”[Mesh]) OR (“Cognitive Dysfunction”[Mesh] OR “Dementia”[Mesh]) AND (“Event-Related Potentials, P300”[Mesh]); Web of Science: TS = (“Alzheimer Disease” OR “Alzheimer’s disease” OR “Cognitive Dysfunction” OR “AD” OR “Dementia”) AND TS = (“Event - Related Potentials” OR “event - related potential” OR “ERP”); Scopus: TITLE - ABS - KEY(“Alzheimer Disease” OR “Alzheimer’s disease” OR “Cognitive Dysfunction” OR “AD” OR “Dementia”) AND TITLE - ABS - KEY(“Event - Related Potentials” OR “event - related potential” OR “ERP”). The search was limited to human participants and publications in English. All articles published before March 2025 were included. All search results were aggregated in Excel for Windows; duplicates were discarded, so unique references were retained at first.

Records were screened according to title and abstracts by two independent researchers. Relevant articles and studies in which eligibility could not be determined based on title or abstract were selected for full-text review. Disagreement between the reviewers was resolved through debate, and the resulting decisions were unanimous.

Studies were eligible for inclusion only if: (I) Participants with AD were diagnosed by formal criteria (McKhann et al., 1984; McKhann et al., 2011); (II) MCI participants were diagnosed according to the Petersen criteria, including only the amnestic types (aMCI) (Petersen et al., 1999); (III) The control group was healthy individuals matched for age and gender; (IV) Used ERPs to detect the relevant indicators, and data on the amplitude and latency of the ERP components (e.g., P300, N400, etc.) were explicitly reported.

Studies with the following conditions were rejected for inclusion: (I) Duplicate publications; (II) Literature for which full text is not available and for which key data are missing; (III) Non-original research literature such as reviews, commentaries, and conference abstracts; (IV) Literature on subjects with other serious neurological or psychiatric disorders that may interfere with ERP results.

Finally, we checked the reference list of included articles, and the references cited within these sources to supplement our database searches, ensuring a comprehensive capture of relevant literature.

The Newcastle-Ottawa Scale (NOS) (Higgins et al., 2011; Wells et al., 2013) was used to assess the quality of the included case–control studies. The scale is scored from three dimensions: the selection of study subjects, comparability between groups, and the measurement of exposure factors, with a full score of 9, and a score of ≥7 is considered high-quality research.

CNV is one of the earliest reported cognitive ERPs, which can detect the arousal level and sustained attention of people under stress, and CNV is a negative potential deflection between the warning stimulus (S1) and the target stimulus (S2), mainly distributed in the frontal region. In the typical CNV task paradigm, S1 and S2 appear in pairs, 30 pairs or more at a time, the subjects need to respond as soon as possible after S2, and the stimulation can be visual, auditory or audio-visual (de Tommaso et al., 2020). The study found that compared with age-matched healthy individuals, there were significant differences in CNV in early AD patients, especially in the late CNV before S2, which may suggest that CNV can provide clues for researching the pathophysiological changes of early AD and may be an important means to identify early AD (Zappoli et al., 1995; Zappoli et al., 1991).

MMN is produced when deviant stimuli are randomly inserted into a series of repeated standard stimuli, reflecting auto-processing of brain to deviant stimuli (Garrido et al., 2009). It originates from the temporal lobe primary and frontal lobe secondary auditory cortex (Sams et al., 1985) with a latency of 100-250 ms. The study found that AD patients have defects in selective attention and working memory capacity, patients are easily distracted by irrelevant information, have a reduced range of memory, decay rapidly, also have executive dysfunction; compared with healthy people, the amplitude of MMN in AD patients is reduced (Tsolaki et al., 2017) and the latency is prolonged (Horvath et al., 2018), reflecting the degree of cognitive impairment in AD higher than in healthy older adults (Chrysa and Papadaniil, 2016; Chrysa and Papadaniil, 2016), indicating that AD patients have impaired attention and reduced auto-processing ability and MMN may be able to predict and identify potential risk groups for AD.

P300 is the third positive wave of ERPs, and its discovery is due to the improvement of the average superposition of neural electrophysiological signals (Friedman et al., 2001). In 1965 Sutton first described the P300: a late positive component that peaks at about 300 ms (Sutton et al., 1965), it consists of three subcomponents, including P3a, P3b, and nP3 (Friedman et al., 2001; Polich and Criado, 2006; Polich, 2007). P300 originates in different regions of the brain, including the temporoparietal junction, the medial temporal lobe complex, and the lateral prefrontal cortex (Huang et al., 2015; Kugler et al., 1993; Briand et al., 2007), P3a reflects frontal function (Polich, 2004), and P3b is associated with the temporoparietal pathway (Huang et al., 2015; Polich, 2004; Warren et al., 2023; Halgren et al., 1980; Fruehwirt et al., 2019). Task-induced P300 processes involve attention, memory, comparison, and decision-making, which constitute the advanced functions of information processing (Hansch et al., 1982). The P300 usually varies with the task intensity and the decision complexity, and its amplitude is also regulated by various factors, such as stimulus significance and resource availability (Chrysa and Papadaniil, 2016). The study found that the P300 latency period prolonged and the amplitude decreased in AD patients compared with healthy people (Duncan et al., 2009; Bobes et al., 2010; Bennys et al., 2007), and the prolongation of the latency period is highly correlated with the severity of cognitive impairment, the more severe the cognitive impairment, the longer the latency period, especially in language, memory, and executive ability (Kostova et al., 2014). Therefore, P300 plays an important role in the evaluation and diagnosis of AD.

The typical N400 task paradigm uses semantically congruent and semantically incongruent stimuli to detect language function. For example, the semantic priming task is the appearance of semantically relevant or irrelevant target stimuli after stimulation with meaningful cues (words, pictures, etc.) (Duncan et al., 2009). When selecting lexical stimulus materials, factors such as word frequency and concreteness need to be considered. The ratio between semantically congruent and semantically incongruent stimuli is often 1:1, and the number of semantically incongruent stimulus trials is ≥40. The amplitude of the typical N400 task paradigm is related to semantic conflict, the stronger the semantic conflict, the larger the amplitude of the evoked N400 (Bobes et al., 2010). It has been found that semantically conflicting paired words can also induce the N400, and the amplitude also becomes smaller as the frequency of stimulus repetition increases (Kostova et al., 2014). The presence of deficits in language and symbol integration processes is one of the typical clinical symptomatic features of AD (Mueller et al., 2018), so N400 may be an indicator of language-related processing deficits in AD. Semantic priming and repetition priming are more frequently used in the study of semantic memory function in AD (Joyal et al., 2020; Perri et al., 2019), and semantic incongruence will cause a larger N400 (Kara and Federmeier, 2005). Compared with healthy older adults, the N400 amplitude of AD patients is smaller, and AD patients have a longer delay of N400 in semantic processing tasks, suggesting that there is a deficit in semantic priming and semantic processing in AD patients (Joyal et al., 2020).

The N100 and P200 are exogenous components related to sensory processing, peaking at approximately 100 ms and 200 ms post-stimulation, respectively. They are associated with attention and their amplitude and latency change with age (Bennys et al., 2007). In auditory tasks involving memory, both the N100 and P200 show increased latency with age (Gmehlin et al., 2011). Research has demonstrated that compared to healthy elderly individuals, AD patients exhibit prolonged latencies of N100 and P200 (Bennys et al., 2007), particularly in response to visual stimulation (Yamasaki et al., 2012; Cecchi et al., 2015). The abnormality of ERPs components in AD patients indicates early-stage impairment of sensory function. These abnormalities reflect differences in cognitive processing between AD patients and healthy elderly individuals, serving as neuroelectrophysiological biomarkers for distinguishing between the two groups.

N170 is a classical visual ERPs component associated with face processing, which amplitude peaks between 140 and 200 ms after face stimulation. N170 is mainly involved in the perceptual process of facial or object feature integration, such as the recognition of human facial identity information (Feuerriegel et al., 2015). One of the most typical clinical symptoms of AD is forgetting familiar faces of family members and companions. This symptom is often attributed to underlying memory impairment (Mazzi et al., 2020), recent studies have found that there are also defects in visual processing (Yamasaki et al., 2012; Fernández et al., 2018; Yamasaki et al., 2012; Fernández et al., 2018; Yamasaki et al., 2012; Fernández et al., 2018). Studies have shown that AD patients have difficulties in color and depth perception, visuospatial organization, control of visual attention, and visual search (Lavallee et al., 2016). The amplitude of N170 evoked by processing facial visual stimuli is reduced in AD patients compared with healthy elderly people, which also suggests that N170 can effectively distinguish healthy elderly people from AD patients, thereby reducing the incidence of AD (Lazarou et al., 2018; Lin et al., 2019). Some studies have found that in the passive observation task using emotional facial expressions, the N170 amplitude of AD patients is significantly lower than that of healthy elderly people (Fide et al., 2019), reflecting the impairment of emotion recognition in AD patients, and impaired emotion recognition may be an early sign of cognitive processing impairment (Fujie et al., 2008). In addition, when only the eyes, nose, mouth and facial contour of the face were presented, the N170 amplitude of the AD patients was also reduced, reflecting the impairment of the ability of perceptual organization during face encoding processing in AD patients (Saavedra et al., 2012). Combined with previous studies, AD patients have not only memory impairment, but also higher-level visual perceptual processing defects, that is, the perception ability of human facial stimulation is decreased, which also suggests that future studies need to use electrophysiological indicators to better predict and find the high-risk groups of AD in the population, so as to reduce the incidence and number of AD (Sabbaghi et al., 2021; Chapman et al., 2011; Eyamu et al., 2023; Quiroz et al., 2011; Polikar et al., 2007; Sabbaghi et al., 2021; Chapman et al., 2011; Eyamu et al., 2023; Quiroz et al., 2011; Polikar et al., 2007).

ERPs have been utilized in the evaluation of cognitive processes, particularly in the early diagnosis of AD and mild cognitive impairment (MCI) (Bennys et al., 2007; Cecchi et al., 2015; Sabbaghi et al., 2021; Vecchio and Määttä, 2011). Studies have demonstrated the potential of ERPs as non-invasive, objective, and low-cost biomarkers for the early detection of AD patients, with high diagnostic accuracy achieved using specific ERP features and classifiers (Sabbaghi et al., 2021; Babiloni et al., 2020). ERPs have been shown to provide powerful and innovative tools for early AD diagnosis, reflecting differences in brain electrophysiology underlying cognitive functions in brain disorders such as dementia and MCI (Eyamu et al., 2023; Bennys et al., 2017).

As a neurodegenerative disease with unknown pathogenesis, the early differential diagnosis of AD is very important. The ERPs component reflects the neurodegenerative process of AD, and its sensitivity holds great promise for detecting and quantifying the presymptomatic stages of AD. Studies have found that AD patients have varying degrees of delay and amplitude changes in each component of ERPs. Early exogenous component changes are the manifestation of sensory impairment, which may be used to distinguish healthy elderly from AD patients. The late endogenous component changes can help to assess the changes in cognitive processes in patients, and the abnormalities of these components can help to differentiate and diagnose the disease. For example, the decrease of N170 amplitude reflects the defect of visual perceptual processing; The prolonged latency of MMN and P300 can be used to detect attention disorders, which is beneficial to the early diagnosis and evaluation of patients; The decrease and delay of N400 amplitude can be used to evaluate language impairment and memory impairment, and track the development of the disease. These components can not only distinguish the healthy elderly from the patients, but also be used to identify the different stages of AD. Therefore, linking ERPs to the diagnosis of the disease could facilitate AD assessment and diagnosis in clinical practice.

ERPs have shown promise in the study of AD, but they also face several limitations. Understanding these limitations is crucial for advancing the field and improving the utility of ERPs in AD research. ERPs signals are usually weak and can be easily affected by background noise. This noise may come from the environment, equipment, or the patient’s own physiological activity (e.g., muscle activity, eye movements, etc.). In order to improve the signal-to-noise ratio, complex data processing and filtering are required, which can affect the accuracy and reliability of the data. ERPs research often requires the design of specific cognitive tasks whose complexity and difficulty may affect patient performance and ERPs response. Different tasks may elicit different ERPs components, and there are large differences in ERPs responses between different individuals, which makes it difficult to draw consistent conclusions in group studies. In addition, factors such as age and gender can also affect ERPs responses, adding to the complexity of data interpretation. While ERPs have high temporal resolution, may not capture very brief neural activity changes in some cases. At the same time, ERPs have low spatial resolution, making it difficult to precisely localize brain activity.

As technology and research methods continue to evolve, ERPs is expected to play a greater role in early diagnosis, disease monitoring, treatment effect evaluation and pathological mechanism research. In the future, higher resolution ERPs equipment can be developed to capture and analyze brain electrical activity more precisely. Using artificial intelligence and machine learning technology, develop more advanced signal processing and filtering algorithms to improve the quality and accuracy of ERPs signals and analysis. Developing multimodal data analysis methods that integrate and interpret data from different technologies to provide a more comprehensive approach to diagnosis and monitoring. ERPs can also be combined with other neuroimaging technologies (e.g., MRI, PET) and biomarkers (e.g., blood and cerebrospinal fluid markers) to gain a deeper understanding of the pathological mechanisms of AD and provide more comprehensive pathological information, thereby providing new ideas and methods for new treatments. At the same time, developing an ERPs-based early warning system to help identify high-risk groups and intervene before symptoms appear. Designing more rigorous longitudinal studies to dynamically monitor changes in cognitive function and assess disease progression, and developing individualized treatment and rehabilitation plans for patients.

Future research should focus on the integration of ERPs with other biomarkers, such as protein biomarkers measured in the cerebrospinal fluid, to improve the differential diagnosis of AD in MCI patients and enhance the predictive models for AD likelihood. There is a need for further investigation into the potential of ERPs, particularly the P300 and N200 components, as preclinical markers of AD, aiming at the earliest possible diagnosis of the disease (Polikar et al., 2007; Bennys et al., 2017). In summary, the meta-analyses and studies on ERPs in Alzheimer’s disease have demonstrated the potential of ERPs, particularly the P300 and N200 components, as sensitive and reliable measures of cognitive deficits associated with early AD. While there are promising applications of ERPs in the early detection of AD, further research is needed to address the limitations and to explore the integration of ERPs with other biomarkers for improved diagnostic accuracy and predictive models.