Increasing empirical evidence supports the notion that affect dynamics - as the temporal fluctuations of our emotional experiences – which is intricately linked with well-being and psychological health (1–3). This burgeoning field of research indicates that not all emotional experiences are created equal, and the way our emotions change and flow over time can provide critical insights into our overall mental health (4–7). In general, medium variability in affective dynamism is associated with good psychophysical health; the opposite effect is observed for extreme states of too high variability or extreme rigidity (e.g., inability to switch from one affective state to another) (8–11). The extreme in affective dynamism is opposite on a descriptive level but equal from a psycho-physiological point of view and have been deeply investigated in the realm of psychiatric or neurological disorders (8, 12). Specifically, affect dynamics have been studied to gain a better understanding of how individuals with conditions such as bipolar disorder, depression, anxiety and eating disorders, experience and feel emotion in time (13–16). In eating disorders, most of all in AN, contemporary theories acknowledge the importance of disordered emotional functioning in the disorder’s development and maintenance [e.g., (17, 18)]. AN is characterized by symptoms of extreme mental and emotional rigidity, making it a compelling case for studying issues in emotion regulation and affect dynamics. Individuals with AN often exhibit a pronounced inability to adaptively manage and transition between emotional states, reflecting significant challenges in emotional flexibility. This rigidity is manifested in both cognitive and emotional domains, leading to a limited range of emotional experiences and heightened difficulty in coping with negative affect (10, 19, 20). Most recent studies have focused on affective variability- pattern of frequent and large mood shifts over time- or on its opposite affective inertia- emotions that are resistant to change or shifts over time (18, 21, 22). The study of Vansteelandt and colleagues (17) investigated affect variability using Experience Sampling Method protocol in patient with AN restricted and Bulimia Nervosa (BN). Results of this exploratory study suggest that the diagnostic groups have the same mean levels of affect, but the AN restricted group showed a less variability in the quality of their affect. Recently, also Williams-Kerver et al. (23) demonstrated, using Experience sampling Method, that participants in the AN and BN groups experienced significantly greater Negative Affect (NA) intensity and better emotion differentiation- the ability to discriminate distinct emotional states- than participants in the Binge Eating Disorders (BED) group. Alternatively, the BN group demonstrated significantly greater NA variability than the AN group and greater NA inertia than the BED group.

All studies described relied most on the Russell’s circumplex model (24), and the predominant experimental design methodology was Ecological Momentary Assessment (EMA) (25). This method allowed collecting real-time data in naturalistic settings, providing a valuable window into individuals’ emotional lives (26–28). EMA involves prompting patients to report on their feelings, behaviors, and environmental context at random times throughout the day, thereby offering a rich, detailed account of affect dynamics (29).

Although EMA is an ecological and longitudinally accurate instrument, it necessitates high participant compliance, that is, the person must be able (and willing) to respond to most of the responses over time. Furthermore, EMA can only delimit and describe the long-term time series of affective states reported by the participant’s responses, assessing mostly the mood; it cannot provide information about short time affective transitions that the person can make during the day or week. Finally, to the best of our knowledge, no one study assesses the physiological characteristic of affect dynamics in eating disorders using EMA, may be due to the compliance and the difficulty of the continuous physiological signals (30–32). Incorporating Heart Rate Variability (HRV), particularly RMSSD, as a physiological signal for affect dynamics is supported by extensive research linking it to autonomic nervous system function and emotional regulation (33). Studies by Thayer et al. (34) and Schmalbach et al. (35) demonstrate that HRV is a robust indicator of emotional regulation and stress resilience, underscoring its relevance for psychological and neurological assessments. Particularly, research by Appelhans and Luecken (36) highlights RMSSD’s specificity in reflecting parasympathetic activity, making it a common measure for estimating the variations in heart rate that are mediated by the vagus nerve and therefore it represents an ideal measure for studying affective dynamics. Recent research underscores the validity of HRV, particularly RMSSD, as a physiological index in psychiatric conditions, linking it to emotional regulation and neurophysiological changes across disorders such as depression and anxiety. HRV reflects the body’s capacity for emotional and physiological regulation, with lower HRV associated with higher symptom severity in psychiatric disorders (12, 35). These findings support the use of HRV in monitoring emotional regulation, offering a rationale for its selection as a physiological index in our study on affect dynamics. Adding RMSSD as a key physiological measure in our study is further justified by its mathematical significance in quantifying short-term variations in heart rate. RMSSD stands for the Root Mean Square of Successive Differences between normal heartbeats. It’s calculated using the formula:

where (RR_i ) represents the duration of one heartbeat to the next (in milliseconds), and (N) is the total number of heartbeats analyzed. This formula captures the variability in heart rate over short periods, making it an excellent index for assessing autonomic nervous system function, particularly the parasympathetic branch. RMSSD’s focus on short-term variability provides a sensitive measure of vagal tone, a key aspect of emotional and physiological regulation.

For this, reason we developed and implemented a new experimental and methodological technique to study AN’s affect dynamics with physiological signals (e.g. RMSSD) in the clinical setting. The goal is to create a simple and effective assessment of affect dynamics, investigating all possible affective transitions and to measure it with Discrete-Time and Discrete Space Markov processes. While previous techniques have focused on group-level data analysis using panel data methodologies, thus, losing sight of the individuality and uniqueness of individual emotional experiences, here, our methodology allows for a personalized, patient-centered perspective. The experimental design involves the emotive images’ visualization of the International Affective Picture System (IAPS), (37) to stimulate all possible affective transitions, based on Russell’s Circumplex model (4). HRV is measured during all the experimental session, and the RMSSD is used into this novel methodological approach of analysis based on Discrete Time and Discrete Space Markovian stochastic processes. Markov chains are mathematical systems that describe a sequence of possible events, where the probability of each event depends only on the state attained in the previous event (38). In the context of affect dynamics, Markov chains can be employed to analyze and predict emotional state transitions over time (39), utilizing physiological measures like RMSSD to understand the probabilistic nature of emotional fluctuations. This approach can effectively capture the dynamic and stochastic nature of emotional fluctuations over time, offering insights into how emotional states evolve (4, 8, 40, 41).

This method stands out for its potential to provide individualized and detailed insights into the emotional patterns of individuals affected by Anorexia Nervosa, offering new perspectives on the relationship between emotions and physiology in these disorders. With this approach, we aspire to develop more effective experimental and methodological tools for monitoring and intervention, enhancing the understanding and treatment of these complex conditions. Through the simple and quantitatively idiographic use of the grapho describing probability transitions, a comprehensive profile can be drawn up that is easy for the patient and the clinician to interpret.

The results are based on the Markovization process, which translates the variability measured by RMSSD into probabilistic terms. The RMSSD is calculated in milliseconds, for each 30’’ transition and for each singular 30’’ self-transition. This approach assesses the participant’s likelihood of variation during transitions between different affective states and the variability within the self-transition (e.g., AA-BB-CC-DD). The Figure 4 presented the value of RMSSD of each transition, before and after the Markovization process.

After a Markovization process, we computed twelve RMSSD transitional indexes (vertical, horizontal, oblique), four static self-transition indexes, and four predictive indexes of steady state. Vertical probability indexes (AC-CA-BD-DB) represent the percentage of transitions between affective states with the same valence levels but opposing arousal levels (high-to-low and vice versa). The horizontal probability indexes (AB-BA-CD-DC) show changes between emotional states that have the same level of arousal but differ in valence (from positive to negative and vice versa). The oblique indexes (AD-DA-BC-CB) indicate the likelihood of transitions between emotional states that involve changes in both arousal and valence. The self-transition indexes (AA-BB-CC-DD) suggest that the participant’s affective state is consistent and unchanging. Steady states represent the likelihood of a person being in a specific emotional state after n step, hypothetically moving between the four quadrants ten steps.

Upon analyzing Figure 5 , discrepancies in the probability of moving from one emotional state to another were noticeable.

Self-transition indexes indicate that the individual with anorexia is more likely to maintain an activated (both positive and negative) emotional state compared to other states ( P _AA = 0.23; P _BB = 0.24; P _CC = 0.13; P _DD = 0.19).

Vertical transitions from high to low activation exhibited a higher probability (for both degrees of valence) than the opposite ones ( P _AC = 0.26; P _CA = 0.13; P _BD = 0.32; P _DB = 0.24).

When transitioning horizontally, there was a clear difference in the probability of moving from positive to negative states compared to the reverse. Particularly, during periods of low arousal, shifts from positive emotions to negative states were more probable ( P _AB = 0.22; P _BA = 0.23; P _CD = 0.15; P _DC = 0.42).

The oblique shifts showed more imbalances in transitioning between the four quadrants, with a greater likelihood of moving from highly activated negative states to low-activated positive states and from low-activated negative states to highly activated positive states ( P _AD = 0.28; P _DA = 0.15; P _BC = 0.21; P _CB = 0.58).

Observing Figure 5 , it is also possible to state that the participant is predicted to experience a significant activation state with positive valence following repeated shifts between emotional states, as indicated by the steady state prediction vector ( P _A = 0.18; P _B = 0.31; P _C = 0.24; P _D = 0.23).

Although the impact of affective dynamics in well-being and health has been studied (62), addressing the physiological layer of this process, also involving an AN patient, poses technical and methodological issues. Our experimental and methodological aim was to develop a patient-centered assessment of dynamism of affective states. First, we assessed the physiological correlate of affect dynamics during the viewing of emotional pictures from the IAPS database, using a laboratory experimental design. Then, using RMSSD indexes into a Discrete-Time and Discrete Space Markov Chain, we were able to compute the probability of transitioning between various emotional states or remaining in a specific state.

Utilizing the Markov chain grapho, a quantitative idiographic approach, we noticed that the patient with anorexia nervosa exhibited particular patterns of behavior. When experiencing lower levels of arousal, the patient was 41% more likely to switch from positive emotional states (such as relaxation) to negative ones (such as boredom), showing strong physiological reactivity. This finding corresponds to other research that emphasizes the difficulties that persons with AN have in regulating their emotions and often encountering heightened negative affect (NA) when exposed to stimuli (18, 63). In conditions of heightened arousal, more balance was observed. This could be attributed to the body’s efforts to maintain homeostasis during stress or engagement, resulting in more regulated emotional responses. Functional MRI studies show that during high arousal, there is increased regulation in brain regions responsible for emotional control, supporting the observation of more balanced emotional states in these conditions (64).

Vertical axis showed increased variability during the shift from high to low activation in conditions with positive valence. Some transitions showed extremely reduced probabilities, particularly when transitioning from low to high activation levels.

As oblique transitions were considered, there was a strong likelihood that the patient would shift from negative to positive emotional states. Observations differed from the outcomes in horizontal transitions, which showed a higher likelihood of moving from positive to negative affective states. To transition from a negative to a positive affective state, the patient needed to regulate their level of activation by switching between low and high arousal. This trend aligns with prior research, which demonstrates the maladaptive techniques for regulating emotions in individuals with AN, such as suppression and avoidance. The need for extreme changes in emotion regulation may suggest the presence of maladaptive methods (18).

The self-transition indexes revealed a greater probability of the patient staying in quadrant B, which represented highly engaged positive emotional states. The likelihood of staying in the other quadrants diminished, resulting in lower variability and reduced sensitivity to these conditions. This observation is consistent with research findings that suggest individuals with anorexia nervosa demonstrate increased functional connectivity in certain brain regions linked to emotional regulation, such as the dorsolateral prefrontal cortex (DLPFC), while simultaneously displaying decreased connectivity in other regions. This disparity adds to their emotional inflexibility and the inclination to persist in certain emotional states for extended durations, thereby reducing the range and responsiveness to shifting emotional circumstances (64, 65).

The steady state matrix displayed a reduced balance among the four quadrants. After ten repetitions of emotional transitions, the patient with AN was more likely to migrate to a highly activated pleasant emotional state, followed by the possibility of experiencing low activating negative emotional states. Once more, our observations align with the findings of Wayda-Zalewska et al. (18), indicating that individuals with AN struggle to maintain psychological balance.

The AN patient exhibited a higher likelihood to engage in oblique transitions, involving changes in arousal and valence. This suggested the requirement of more opposing emotional inputs to facilitate a shift from one emotional state to another. The variability indexes of the patient with anorexia nervosa were diversified: her indexes exhibited substantial variation in some cases and were nearly absent in others, revealing various physiological response patterns based on the emotional transition’s characteristics.

Emotional difficulties in AN include inappropriate emotion regulation (ER) and expression (66), increased negative affect (NA) in response to unpleasant stimuli (63), and alexithymia (67). Among various emotional experience patterns, NA is critical for the development and maintenance of AN. Studies show that higher daily levels of NA increase the likelihood of food restrictions. NA can also increase following specific AN behaviors such as loss of control (LOC) eating, purging, a combination of LOC and purging, or weight checks. Conversely, NA levels tend to fall after drinking fluids or engaging in physical exercise. Research has focused on the states of NA and their specific manifestations (e.g., anxiety or tension) in AN, along with behaviors such as dietary restriction, LOC, or purging, as well as frequent weighing (18).

To contribute to the advancement of affect dynamics’ assessment research, we introduced a novel experimental and methodological approach that examines cardiac variability, specifically RMSSD, in a clinical and laboratory setting. (68, 69). Using standardized IAPS emotional imagery, researchers can effectively investigate affective transitions in a controlled manner, allowing the detection of cardiac variability parameters.

Using an index like RMSSD, one of the most important indicators for cardiac variability, we could evaluate the short-term variability of the various affective transitions. Furthermore, employing Discrete Time and Discrete Space Markov chains enables us to statistically translate the variability associated with transitions into indexes that show the likelihood of variation between states. The Markov matrix measures descriptive indexes—those calculated on the transition matrix, such as verticals, obliques, horizontals, and trait states—as well as predictive indexes—such as steady states—resulting from the matrix product of the transition matrix and the initial state vector (39). Descriptive indexes provide a snapshot of the patient’s current affective state, while predictive indexes forecast the likelihood of being in one of four affective states after multiple transitions. Empirically, we’re curious about how the participant reacts to an emotional storm caused by the encountered emotional shifts. Will the patient feel stressed, relaxed, or engaged?

The combination of an experimental design with analytical techniques like Markov matrixes creates a patient-centric approach, providing detailed insights into a patient’s affectivity, complementing for a holistic understanding of a patient’s emotional health. Our findings suggest that understanding affect dynamics through a combination of Markov Chains and HRV might improve current therapeutic approaches for treating AN. This would enable the implementation of an emotional assessment that is experimentally simple to measure and clinically simple to interpret for both the clinician and the patient. Indeed, the graph of transitions and steady states are quantitative idiographic methods that provide an instant comprehension of the patient’s emotional dynamism.

The current study has limitations by its nature as a case study, which restricts the generalizability of the observed results. Further studies should employ larger and more varied samples to improve generalizability. Extending this methodology to people with differing severities of AN, comorbidities, or other disorders—such as mood or anxiety disorders—necessitates consideration of condition-specific variations in autonomic and emotional control. Additionally, the absence of assessments on conscious emotional processing, such as explicit questions, hinders our understanding of the specific emotions acknowledged by patients. It would be intriguing to use this component in future research to detect any alignment or disparity between the implicit and explicit regulation aspects. Furthermore, self-report surveys assessing the individual’s emotional regulation styles could be implemented.

Future directions also involve incorporating virtual reality (VR) (70–72) to create dynamic stimuli that can better replicate the transitions and emotional dynamics experienced by individuals in the real world. Future research will also explore the idea of creating tools for the generation and examination of Markov matrixes.