OUP user menu

Believing is perceiving: mismatch between self-report and actigraphy in psychogenic tremor

Isabel Pareés , Tabish A. Saifee , Panagiotis Kassavetis , Maja Kojovic , Ignacio Rubio-Agusti , John C. Rothwell , Kailash P. Bhatia , Mark J. Edwards
DOI: http://dx.doi.org/10.1093/brain/awr292 117-123 First published online: 10 November 2011


We assessed the duration and severity of tremor in a real-life ambulatory setting in patients with psychogenic and organic tremor by actigraphy, and compared this with self-reports of tremor over the same period. Ten participants with psychogenic tremor and eight with organic tremor, diagnosed using standardized clinical criteria, were studied. In an explicit design, participants were asked to wear a small actigraph capable of continuously monitoring tremor duration and intensity for 5 days while keeping a diary of their estimates of tremor duration during the same period. Eight patients with psychogenic tremor and all patients with organic tremor completed the study. Psychogenic patients reported significantly more of the waking day with tremor compared with patients with organic tremor (83.5 ± 14.0% of the waking day versus 58.0 ± 19.0% of the waking day; P < 0.01), despite having almost no tremor recorded by actigraphy (3.9 ± 3.7% of the waking day versus 24.8 ± 7.7% of the waking day; P = 0.001). Patients with organic tremor reported 28% more tremor than actigraphy recordings, whereas patients with psychogenic tremor reported 65% more tremor than actigraphy. These data demonstrate that patients with psychogenic tremor fail to accurately perceive that they do not have tremor most of the day. The explicit study design we employed does not support the hypothesis that these patients are malingering. We discuss how these data can be understood within models of active inference in the brain to provide a neurobiological framework for understanding the mechanism of psychogenic tremor.

  • actigraphy
  • movement disorder
  • psychogenic
  • tremor


Psychogenic tremor is the most common form of psychogenic movement disorder (Edwards and Schrag, 2011). In common with other psychogenic movement disorders, and indeed psychogenic motor symptoms in general, the clinical features that are used on examination to distinguish such patients from those with organic motor disorders rely primarily on demonstrating a mismatch between symptoms as presented by the patient and symptoms when the patient is distracted by particular examination manoeuvres. For example, in the case of psychogenic tremor, a distinct change in the tremor is typically observed when patients are asked to do a sudden ballistic movement with the other hand, or to perform complex sequential movements with fingers of the contralateral hand (Kumru et al., 2004, 2007). While these effects are also observed in healthy participants mimicking tremor, they are not typically seen in patients with well defined organic tremor disorders such as Parkinson's disease or essential tremor. Recent functional imaging studies underline the similarities between voluntary movement and psychogenic tremor (Voon et al., 2010).

Despite the similarity between the clinical features of psychogenic tremor and voluntary tremor, patients with psychogenic tremor subjectively report their tremor as involuntary and persistent, often causing severe disability (Jankovic et al., 2006). Psychogenic movement disorders are commonly considered to be a form of conversion disorder and only rarely thought to be due to deliberate production of symptoms for psychological (factitious disorder) or personal (malingering) gain. Typically, psychological stressors are implicated in the origin of psychogenic disorders, but the evidence for this is limited (Binzer and Eisemann, 1998; Roelofs et al., 2002, 2005), and even if this hypothesis is correct, the neurobiological mechanism by which psychological stressors could cause such unusual physical symptoms is unknown.

In this study, we took advantage of the ability to assess the presence and intensity of tremor accurately, remotely, and for long periods of time using a wrist-worn actigraph device. In contrast to cumbersome devices used in the past for ambulatory tremor monitoring (Spieker et al., 1995, 1998), this device is small and has been demonstrated to accurately differentiate tremor from other movements (Van Someren et al., 2006). We used this device in a cohort of patients with psychogenic tremor and organic tremor to determine the duration and intensity of tremor in a natural setting over 5 days, and compared these data with self-report of tremor duration over the same period and a standardized face-to-face clinical assessment of tremor severity.

Materials and methods


We recruited 10 patients with psychogenic tremor from movement disorder outpatient clinics at the National Hospital for Neurology and Neurosurgery from March 2010 to January 2011. Inclusion criteria were age >18 years, a diagnosis of clinically established or documented psychogenic tremor according to Fahn and Williams criteria (Fahn and Williams, 1988) and tremor in at least one arm at rest, on posture or both of a moderate/severe level judged by a score of at least 2 on Part A of the Fahn–Tolosa–Marin scale (Fahn et al., 1988). Patients with any major concurrent neurological disorder were excluded. Eight patients with different forms of organic tremor served as a control group. They all had clinically typical tremor and course of illness for their diagnosis and had to suffer moderate/severe tremor (at rest, on posture or both) in at least one arm judged by a score of at least 2 on Part A of the Fahn–Tolosa–Marin scale. Patients with marked clinical fluctuations in response to medication were not included.

At the inclusion visit, tremor was rated using the Fahn–Tolosa–Marin scale, which has three subscales (A: tremor location and severity; B: specific motor tasks and function; and C: functional disabilities due to tremor). Participants were videotaped during clinical assessment. Hand dominance was assessed by the Edinburgh Handedness Inventory (Oldfield, 1971) and EuroQoL-Five Dimensions (EQ-5D)™ (Brooks et al., 2003) was used to assess quality of life. The study was approved by the National Hospital for Neurology and Neurosurgery/Institute of Neurology Joint Research Ethics Committee. The purpose of the study was explicitly explained to the participants and a written consent to take part in the study was obtained according to the Declaration of Helsinki. A post-study questionnaire was designed to assure that participants understood the purpose of the study. Patients were retrospectively contacted by telephone and were asked two questions: (i) ‘What do you think was the purpose of wearing the watch?’; and (ii) ‘For how much of the time that you were wearing the watch do you think it was turned on?’

Tremor recording

The Actiwatch (Cambridge Technology) contains a uniaxial accelerometer consisting of a small mass fixed to a piezoceramic bar. When the piezoceramic bar is distorted by acceleration, a current is induced that is proportional to acceleration. The device continuously samples the output of this internal accelerometer at 64 Hz, with an 8-bit resolution covering −5 to +5G. The algorithm programmed in the Actiwatch has been validated to discriminate tremor from other movements with high sensitivity and specificity (Van Someren et al., 2006). Continuous recording of duration and intensity of tremor for up to 22 days is possible. Optimal sensitivity is achieved for tremor of a frequency ≥3 Hz (Van Someren et al., 2006; Binder et al., 2009). Tremor duration is reported by the Actiwatch software as seconds of tremor per minute of recording. Tremor intensity is reported as the highest amplitude of tremulous movement in each minute measured as counts (25 counts/s representing approximately an acceleration of 1G; The Actiwatch User Manual, version 7.2; http:www.camntech.com).

Participants were instructed to wear the actigraph on the wrist of the most tremulous arm constantly for five consecutive days. They were instructed that the actigraph should only be taken off when the hands were exposed to water (e.g. showering or swimming).


Self-rated assessments regarding the tremor for the same days as the actigraph was worn were reported in a self-completed diary. This had five sheets, each sheet covering 1 day. On each sheet participants were asked to record the time of waking up and going to bed, as well as any time that they spent without wearing the actigraph. For three predefined intervals per day (time from waking to 1 p.m.; from 1 p.m. to 7 p.m.; and time from 7 p.m. to bedtime) they were asked to mark the percentage of time they estimated themselves as having tremor on a visual analogue scale marked from 0 (no tremor) to 100 (tremor 100% of the time interval). We termed these ‘intra-day interval estimations’. We also asked participants to estimate on average the proportion of the whole waking day that they had experienced tremor by using the same type of visual scale. We termed this the ‘whole day period estimation’. Finally, we asked subjects to record whether they felt that the day had been typical for them in terms of the amount of tremor they experienced.

Data analysis and statistics

Duration of the waking day was defined as the time from waking up to bedtime minus the amount of time that each participant took the actigraph off in minutes. Duration of tremor was calculated as the total amount of seconds with tremor as measured by the actigraph during the waking day. The results were expressed both in minutes and as percentage of the duration of the waking day. To compare intra-day interval estimations reported in the daily diary with the results of the actigraph, we converted the amount of time with tremor measured by the actigraph to a percentage of each interval of time.

We used the PASW statistical package (version 18) and MedCalc (version 11.6) for statistical analysis. P-values for categorical variables were calculated with the use of Fisher's exact test. T-tests were used to compare differences in means for numerical data when parametric assumptions were met and Mann–Whitney U-tests when data were not normally distributed. Statistical significance of P < 0.05 was assumed. Bland–Altman analysis corrected for repeated measurements was used to assess the agreement between ‘intra-day interval estimations’ and tremor measured by actigraphy (Bland and Altman, 2007). Results are expressed as the mean bias (difference between tremor reported in diaries minus tremor recorded by actigraphy: positive values indicating overestimation in diaries and negative values indicating underestimation in diaries) and 95% confidence intervals (CIs). We plotted these against the geometric mean of the two measures (calculated by multiplying ‘intra-day interval’ estimations and results from the actigraphy, and taking the n-th root, where n was the number of values to average), giving us an opportunity to assess bias when diary and actigraphy scores were at different levels. Intra-day interval estimations not clearly reported were excluded from this analysis.


Baseline characteristics

Eight patients with psychogenic tremor and eight patients with organic tremor completed the study. After initially agreeing to take part, two patients with psychogenic tremor decided not to complete the study. One reported severe pain in the affected arm meaning that he could not wear the watch and the other reported an allergic skin reaction due to the watch strap. In the group with organic tremor, five (62.5%) patients had Parkinson's disease, two (25%) had dystonic tremor and one (12.5%) had Wilson's disease.

Table 1 summarizes the clinical and demographic data. The group with organic tremor was slightly older than the group with psychogenic tremor (66.1 years versus 52.3 years; P = 0.0054), but they were matched on tremor severity as judged by the Fahn–Tolosa–Marin scale. Two (25%) psychogenic patients and six (75%) organic patients had tremor elsewhere apart from the arm wearing the actigraph. Patients with psychogenic tremor rated their quality of life as significantly more impaired than patients with organic tremor. Two (25%) psychogenic patients were receiving treatment for tremor at the time of the study (propranolol in both cases), whereas six (75%) organic patients were on treatment (anti-parkinsonian drugs in five cases and propranolol in one case).

View this table:
Table 1

Baseline characteristics of the patients

Demographic and clinical informationPsychogenic tremor (n = 8)Organic tremor (n = 8)P-value
Age, mean (SD) (years)52.3 (12.1)66.1 (14.1)0.054
Sex, n (%)
    Male4 (50)6 (75)0.61
    Female4 (50)2 (25)
Handedness, n (%)
    Right6 (75)7 (88)1.0
    Left2 (25)1 (12)
Disease duration, mean (SD) (years)10.7 (11.8)7.6 (4.7)0.92
Triggering event prior to onset of tremor
    Yes6 (75)00.007
    No2 (25)8 (100)
Fahn–Tolosa–Marin tremor scale, mean (SD)
    Subscale A9.1 (3.7)12.5 (7.8)0.29
    Subscale B9.5 (6.0)10.6 (4.8)0.69
    Subscale C15.2 (4.4)8.1 (5.4)0.01
    Total34.0 (11.3)31.1 (13.8)0.66
EQ-5D (dimensions), n (%)a
    Mobility8 (100)3 (37.5)0.03
    Self-care8 (100)1 (12.5)0.001
    Usual activities8 (100)5 (62.5)0.2
    Pain7 (88.5)00.001
    Anxiety/depression6 (75)2 (25)0.13
EQ-5D visual analogue scale, mean (SD)42.6 (22.0)75.4 (8.0)0.002
Treatment for tremor, n (%)
    No6 (75)2 (25)0.13
    Yes2 (25)6 (75)
Active compensation/litigation00
  • a Number of patients (%) reporting some/extreme problems.

Comparison between actigraphy and self-report of tremor

There were no differences between groups regarding their reported duration of the waking day and the amount of time they were not wearing the actigraph (Table 2). Patients with psychogenic tremor reported the 5-day period of the study as 100% representative of a normal day in terms of tremor, whereas patients with organic tremor considered 87.5% of the 5-day period as representative. None of the actigraphs ran out of power during the experiment.

View this table:
Table 2


Assessed variablesPsychogenic tremor (n = 8)Organic tremor (n = 8)P-value
Waking day duration, mean (SD) (min)860.8 (222.6)975.6 (47.3)0.93
Time without the actigraph, mean (SD) (min)36.2 (25.7)24.4 (21.4)0.42
No. of representative daysa, mean (SD)5 (0)4.4 (1.1)0.64
Tremor intensity, mean (SD) (counts/s)13.6 (10.25)12.2 (5.84)0.83
Waking day tremor duration
    Diary, mean (SD) (%)83.5 (14.0)58.0 (19.8)0.01
    Actigraphy, mean (SD) (%)3.9 (3.7)24.8 (7.7)0.001
    Actigraphy, mean (SD) (min)31.1 (30.7)240.0 (70.1)0.001
  • All variables are the average values of the five waking days.

  • a Number of days that participants reported as representative of a normal day in terms of tremor.

In diaries, ‘whole day period’ estimation for duration of tremor, as percentage of the waking day, was significantly higher in patients with psychogenic tremor than patients with organic tremor (83.5 ± 14.0% versus 58.0 ± 19.0%, respectively; P < 0.01). However, mean percentage of the waking day with tremor measured by actigraphy was significantly lower in the psychogenic tremor group (3.9 ± 3.7%; 31.1 ± 30.7 min) compared with the organic tremor group (24.8 ± 7.7%; 240.0 ± 70.1 min) (P = 0.001). The mismatch between diary estimates and actigraphy measures is illustrated in Fig. 1. When tremor was present, there was no difference between groups in the intensity of tremor measured by the actigraph (P = 0.83).

Figure 1

Tremor duration as percentage of the waking day [mean (SD)], as recorded in self-report diaries and by actigraphy, in patients with organic tremor (OrgT) and psychogenic tremor (PsyT).

Since both patients with psychogenic tremor and organic tremor overestimated the amount of time with tremor in the ‘whole day period’ estimations, we analysed the agreement between diary and actigraphy measures in more detail by assessing each of the ‘intra-day interval’ estimations as recorded in diaries and actigraphy over the 5 days (Fig. 2). Bland–Altman analysis showed a significantly better agreement between methods in the group with organic tremor compared with the group with psychogenic tremor (organic tremor: mean bias = 27.6%, 95% CI: 26.0–81.2; psychogenic tremor: mean bias = 64.7%, 95% CI 1.2–128.1). According to this analysis, intra-day interval estimations of patients with organic tremor showed 28% more tremor duration in diaries than by actigraphy. In contrast, patients with psychogenic tremor showed 65% more tremor duration in diaries than by actigraphy. The fairly random spread of data points in Fig. 2A and B, with regard to the x-axis, indicates that bias was not proportional to the mean of tremor duration estimates and actigraphy, but rather subjects had an absolute systematic bias towards overestimation of tremor.

Figure 2

Bland–Altman plots for tremor duration as percentage of the intra-day intervals, recorded in self-report diaries and by actigraphy, in patients with organic tremor (A) and psychogenic tremor (B). Symbols in Fig. 2A and Fig. 2B each refer to individual values for separate patients (those with organic tremor in Fig. 2A, those with psychogenic tremor in Fig. 2B). Bias is expressed by the mean of the differences between methods and 95% CIs (±1.96 SD). A difference of 0 (dotted grey line) represents the perfect agreement between both methods. Differences with positive values indicate an overestimation of tremor duration in diaries compared with actigraphy. Differences with negative values indicate an underestimation of tremor duration in diaries.

We additionally assessed intensity and duration of tremor at the time when patients were supposed to be filling in the diaries. Although we gave them instructions to fill in the diary at particular times of the day, we cannot be completely sure that they strictly followed the instructions. The most reliable moment that we believe we can be certain that the patients were filling the questionnaire was at bedtime. We therefore analysed the intensity and duration of tremor during the 30 min before going to bed. We did not find any increase in the intensity of tremor during this period in either group. However, for tremor duration, when we calculated the ratio between the percentage time with tremor over the 30 min before going to bed compared with the percentage time with tremor over the whole third interval (from 7 p.m. to bedtime), we observed significant differences between groups (P = 0.01). The group with psychogenic tremor had a ratio of 1.7 (indicating an increase of tremor during the 30 min before going to bed with respect to the rest of the interval), whereas the group with organic tremor had a ratio of 0.7 (indicating a decrease of tremor in the 30 min before going to bed).

In the post-study questionnaire, all patients answered that they believed that the purpose of the study was to monitor tremor, and that the watch was recording tremor 24 h a day.


In this study, we have assessed duration and intensity of tremor in patients with psychogenic tremor and organic tremor by using a validated long-term actigraph during 5 days and simultaneous self-rated measurements. We have demonstrated a remarkable absence of tremor during most of the waking day in patients with psychogenic tremor. Despite this, patients with psychogenic tremor reported tremor to be present majority of the time. In comparison with organic tremor group, they reported tremor to be present in a significantly higher proportion of the waking day, and when compared with the results of actigraphy, showed a significantly greater bias towards over-estimation of tremor.

Patient groups were well matched in terms of objective clinical assessment of tremor severity at baseline. In line with their estimates of tremor duration, but not with the actigraphy data, patients with psychogenic tremor rated themselves as significantly more disabled by tremor (subscale C of the Fahn–Tolosa–Marin scale), and as having a significantly poorer quality of life as rated by the EQ-5D.

Implications for understanding the pathophysiology of psychogenic tremor

We suggest that these data provide evidence that the majority of patients with psychogenic tremor (and perhaps by inference other psychogenic disorders) are not malingering. We were explicit in our explanation of the study to the patients, specifically explaining that we were using a ‘tremor watch’ to record the actual duration of their tremor and were comparing this with patients’ own estimates of tremor duration. We gave patients the opportunity to abort the study at any point. Two of the patients with psychogenic tremor initially accepted our invitation to take part in the study, but then returned the actigraph to us without wearing it or completing diaries. The reasons given for this (arm pain, allergic reaction) may be genuine, but could also be hypothesized to be excuses to avoid revealing that tremor was not present when unobserved, compatible with the diagnosis of factitious disorder/malingering. However in the rest of the psychogenic tremor group, patients completed the study in a similar fashion to organic patients. This behaviour, given the explicit study design, seems incompatible with malingering or factitious disorder, and instead suggests a genuine perception that tremor was present to the extent reported in the diaries.

If malingering/factitious disorders are not explanations for the tremor in our group of patients, how might these data inform our understanding of the mechanism of symptom production? These data reflect the interaction between perception (diary assessments) and sensory data (actigraph measurements). In an ideal neural model these would be identical (i.e. all sensory data are correctly translated into perceived tremor), but we know from personal experience and numerous experimental studies that perception of sensory data is dramatically altered by expectation (Koyama et al., 2005; Colloca et al., 2008; Bulsing et al., 2010). For example, perception of pain can be radically altered by expectation of the intensity of the pain stimulus (Koyama et al., 2005; Colloca et al., 2008).

This interaction between expectation and sensory data is the basis of models of predictive coding (or more generally hierarchical Bayesian inference) and constitutes a fundamental cognitive principle of global brain function. According to these theories, the brain is an inference machine that actively predicts and explains its sensations (Friston, 2010). Perception, thus, arises from integrating sensory information from the environment with internal predictions about the expected data (priors). Mismatch between sensory information and priors is known as prediction error. In physiological states, neuronal systems are hypothesized to try to minimize prediction error through interactions between multiple levels of a cortical hierarchy (Friston, 2010). The minimization of prediction error can occur via a variable combination of modifications to priors so that they better fit the sensory data and by actively adjusting the precision that is assigned to sensory data to better conform to priors. This process treads a fine line between the neural system reacting chaotically by changing priors to fit any incoming sensory data regardless of its importance and maintaining predictions so immutably that the response of the organism is entirely insensitive to changes in the environment (Corlett et al., 2009).

We suggest that our data fit within this framework, and specifically that prior expectations about tremor in both psychogenic and organic patients influence the way they estimated their tremor. Both patient groups overestimated their time with tremor, and this could be conceptualized as an overweighting of prior expectancy about having tremor over actual sensory data. Such overvaluing of prior expectancy is indeed the norm in most studies of perception in healthy populations, and may reflect an evolutionarily beneficial tendency to value past experience over new sensory information (Elze et al., 2011). However, our data, with regard to the significantly greater estimation/actigraph mismatch in patients with psychogenic tremor, fit with an abnormal weighting of prior expectancy in this patient group such that it overwhelms sensory data that should inform the patient that they do not have tremor. We hypothesize that whenever patients’ attention is turned towards the symptom (for example, during clinical examination or as demonstrated in the present study, around the time they had to fill in the diaries) the expectation of the sensory consequences of the symptom is of sufficient strength to drive the abnormal motor behaviour. On the contrary, when attention is diverted away from the symptom (for example, during most day-to-day behaviour away from the clinic) tremor stops. However the patient's perception, which is moulded by the dominance of the abnormally strong prior expectation over sensory data, remains that tremor is present most of the time.

Implications for clinical trials in psychogenic tremor

This is the first study to assess psychogenic tremor objectively in a ‘real-life’ ambulatory fashion. To date, information regarding severity of symptoms in psychogenic tremor has been provided by objective face-to-face clinical observation using standardized rating scales such as the Fahn–Tolosa–Marin scale or from patient self-report (Jankovic et al., 2006; McKeon et al., 2009). Our data demonstrate that although intensity and severity of psychogenic tremor, when present, is similar to organic tremor measured by actigraphy and rated using standardized scales, patients’ self-reports do not capture the reality of tremor duration in day-to-day life. This finding has implications for design of future clinical trials. In essence, our data suggest that the disability reported by patients with psychogenic tremor is not due to the tremor itself, but more to their abnormal perception that the tremor is continually present. This would argue for the use of global disability and quality of life measures along with specific tremor assessments, as outcomes in clinical trials in this condition.

Study limitations

We acknowledge some limitations to our study. First, we have studied a small cohort of psychogenic and organic tremor patients, and we cannot exclude that in a larger cohort data may be different. However, we chose patients with clinically typical psychogenic and organic tremor diagnosed using standardized criteria and feel that they do accurately represent patients with these diagnoses. We cannot rule out that the actigraph was underestimating tremor in both groups, as in a previous study 71% of 10 min of tremor observation were classified as tremor by the actigraph (Van Someren et al., 2006). However, studies using EMG, which is more sensitive to tremor than actigraphy (Spieker et al., 1997), have found patients with Parkinson's disease to have tremor 28.9% of a 24-h period and patients with essential tremor to have 15.8%. These results are similar to our data in the organic tremor group. Also, we cannot completely exclude the possibility that tremor was not accurately recorded by actigraphy in patients with psychogenic tremor because of the known variability in psychogenic tremor frequency. However, this seems to be unlikely since the filters for tremor were set over a wide range (3–11 Hz). Even though participants were clearly instructed, we cannot exclude that they overestimated tremor duration because tremor involving other body parts was also reported. Nevertheless, the majority of patients with psychogenic tremor had tremor only in the arm wearing the actigraph, and consequently, this explanation is unlikely to account for the excessive overestimation of tremor in the psychogenic tremor group. Finally, we acknowledge that assessment of the presence or absence of malingering is very difficult and our data cannot fully exclude this possibility, and therefore that there was some purposeful embellishment in the way psychogenic patients completed the diaries. However, a post-study questionnaire indicated their understanding of the nature of the study and we feel malingering is an unlikely explanation of our results.


We have demonstrated a dramatic overestimation of duration of tremor in patients with psychogenic tremor when compared with estimates of patients with organic tremor and ambulatory actigraphy. These data fit with a neurobiological explanation for psychogenic tremor based on an abnormal overweighting of prior expectancies regarding tremor, which might distort perception. Our data do not support the hypothesis that these patients are malingering. While speculative, the conceptualization of psychogenic disorders as disorders of active inference allows the generation of testable hypotheses regarding the pathophysiology of these common, disabling and poorly understood disorders.


This work was performed at University College London Hospitals/University College London who received a proportion of funding from the Department of Health's National Institute for Health Research Biomedical Research Centres funding scheme. National Institute for Health Research Research Training Fellowship (to T.A.S.); National Institute for Health Research Clinician Scientist Fellowship (to M.J.E.).


We thank Dr Zapater and Dr Horga for their helpful comments on the manuscript.


View Abstract