We used a cross-sectional design with one measurement timepoint. Participants, recruited at GHU Psychiatrie et Neurosciences hospital (Paris, France), signed an informed consent. The study was approved by the ethical review board (CPP Sud Mediterranee V nr: 2018-A01945-50, PI MO Krebs).

This pilot sample consisted of four groups, each with N = 20 participants, were included: a Control, FEP, SCZ and ASD group. The control, SCZ and ASD groups were age-matched, but the FEP group was younger, given that FEP is typically discovered prior to clinical diagnosis of schizophrenia (Table 1). The diagnoses were made by A.L., G.T, and V.M. before study inclusion.

The inclusion criteria were:

For all groups: age > 18 and < 50 years (except for FEP), be affiliated to a National Health Insurance (or equivalent) scheme.

For patients with SCZ: be diagnosed with schizophrenia or schizoaffective disorder under DSM-V criteria. Stable symptomatology; treated with a stable dose of antipsychotics for >1 month.

For patients with FEP: >18 and < 30 years old, criteria for psychosis according to Comprehensive Assessment of At Risk Mental States (CAARMS), within 2 years.

For patients with ASD: diagnosed using DSM-V criteria.

The exclusion criteria were:

An IQ < 70 [assessed by WAIS-IV (45)]. Standard contraindications to TMS [as in (30)]. For healthy controls: any current or previous psychiatric disease or any familial psychiatric background at the first degree.

Clinical assessments: all patients were assessed using NSS (18), Positive and Negative Syndrome Scale (PANSS) (46), Brief Psychiatric Rating Scale-24 (BPRS) (47), and the Stroop test (48). The NSS scale covers 25 items divided into four subcategories: Motor Coordination (MoCo), Sensory Integration (SI), Motor Integration (MI), and Involuntary Movement. Clinical exams were followed by manual dexterity and TMS assessments.

The tablet application tasks were developed in our lab in collaboration with Sensix. ¹ The tasks were inspired from clinical NSS procedures and previously developed measures of dexterity (49). Validity and reliability of these tablet tasks has been studied previously in healthy subjects (39). During the tablet tasks, the subject’s attention was directed to control of the fingers and use of visual/auditory task instructions and feedback. If a finger was positioned outside of the area where fingertip contacts and movements were detected, the task automatically stopped and restarted again when the fingers were correctly replaced. This procedure helped to control for drifts of attention. Furthermore, task duration was limited and pauses were inserted between tasks to reduce the risk of distraction and fatigue.

All tablet tasks were performed twice in a standardized manner (39), at least 1 h apart: first for familiarization, then for assessment. Performing all five tasks took ~20 min. The fingers must be placed and maintained on the screen during the whole task: if they moved too far, the tasks stopped. Each task is briefly described below.

Finger Recognition (FR): This task was designed to assess the ability for effector selection requiring mental rotation. A virtual hand appeared on the top half of the screen, opposite to the subject’s hand (Figure 1A inset). The subject had to produce a single tap, i.e., extension and flexion in a single movement, with the finger corresponding to the virtual target finger, as fast as possible and without moving other fingers (referred to as coactivations). This task consisted of three conditions: (I) Mirror condition: target and tapping finger were vertically aligned (Figure 1A inset), requiring no mental rotation. (II) Inverse condition, requiring mental rotation: the virtual hand, oriented vertically, was left/right inverted. (III) Perpendicular condition, requiring another type of mental rotation: the virtual hand was oriented horizontally (90° rotation). 30 trials (finger taps) were performed per condition, with 6 blocks (2 Mirror, 2 Inverse, 2 Perpendicular) of 5 randomized trials each.

Performance variables: reaction time, correct effector activation (% trials), and coactivation rate in non-instructed fingers (% trials).

Rhythm Tapping (RhT): This task was designed to assess temporal control of finger movements. The subject had to repeatedly tap with a single finger in synchrony with an auditory cue at a given constant (1, 2 or 3 Hz) frequency (Figure 1B inset). This Feedback condition was immediately followed by a NoFeedback condition, where the subject had to continue tapping at the same frequency without auditory cue. This was repeated separately for each finger.

Performance variables: mean inter-tap-interval (ITI), intra subject variability (ISV; the standard deviation of ITI).

Multi Finger Tapping (MFT): This task was designed to assess finger selection and individuation. The subject had to perform a single-finger tap or a simultaneous two-finger tap following a visual cue, as fast as possible and without moving non-target fingers. 30 single-tap trials (6 trials per finger) and 60 two-finger tap trials (6 trials per two-finger combination) were performed in randomized order.

Performance variables: reaction time, correct activations (% trials), and coactivation rate in non-instructed fingers (% trials).

Sequence Tapping (ST): This task was designed to assess the ability to perform and memorize motor tapping sequences. A sequence of 5 single-finger taps was displayed on the screen and the subject was instructed to tap to each cue with the respective finger. After 10 repetitions, the subject had to repeat (5 times) the learned sequence without visual cues. Sequence_A consisted of [3-4-2-5-1] (1 = thumb, 5 = little finger) and sequence_B of [4-3-5-2-1].

Performance variables: mean number of successful taps within a trial (Success Taps Trials, range: 0–5).

Line Tracking (LT): This task was designed to assess visuomotor integration and attentional capacities. The subject had to follow a curved unpredictably moving line on the screen with the index finger (Figure 1C inset). A dual-task paradigm was also implemented by adding a cognitive task: during line-tracking, shapes or single numbers (1–9) appeared briefly on the screen. The subject was instructed to ignore shapes (distractors), but to mentally subtract numbers successively from 50. Details in Supplementary methods.

Performance variables: total task duration (which depended on tracking accuracy and velocity), mean number of velocity peaks per second extracted from the entire trajectory, and number of micromovements (noise) during the task.

Tablet performance was analyzed with MATLAB ⁶ and statistical analysis was performed with SPSS ⁷ and R ⁸. Normal distribution of data was assessed through Shapiro–Wilk tests. The following tablet measures were normally distributed: FR and MFT reaction times, and RhT IntraSubject Variability (ISV). TMS measures were normally distributed. Normally distributed variables were analyzed with ANOVAs with factors GROUP (Control, FEP, SCZ, and ASD), CONDITION (task-related) factors, and the CONDITION*GROUP interaction, including Bonferroni correction for post-hoc tests. RhT ISV was also analyzed for CUE and for GROUP*CUE interaction. Skewed variables were analyzed with non-parametric Scheirer–Ray–Hare tests (52, 53) with the same factors, and Dunn’s post-hoc test (corrected for multiple comparisons). The level of significance was set to p < 0.05. Radar plots with z-scores for each group were calculated to summarize performance results.

Correlations between tablet performance variables and age, medication, NSS outcomes, and other clinical scales were tested with Spearman’s rank-order correlation.

Receiver Operating Characteristic (ROC) curves were used to evaluate group discrimination according to area under the curve (AUC), sensitivity and specificity of three dexterity measures (which showed the most significant group differences), and of NSS. We tested FEP vs. controls (One-vs-One approach), then FEP vs. controls, SCZ and ASD (One-vs-All approach). Cut-off points were identified with the Youden Index.

A decision tree analysis based on tablet performance was computed to classify the subjects according to groups. Data from our previous study (39) of 9 healthy control subjects was added to this analysis to increase statistical power. These subjects were examined using the same protocol, but without TMS measures. Twelve performance variables with most significant group differences, three from each of four tablet tasks (FR, RhT, MFT, and LT), were used for the decision tree analysis. A complementary Random forest analysis was performed, with training and validation samples (additional information in Supplementary methods).

All groups (except FEP) comprised slightly more men than women. Patients with FEP were younger; the other groups were age-matched. Age of first episode was lowest in ASD, and equivalent in FEP and SCZ. Medication was similar among patients. This and symptom scores are detailed in Table 1.

In the Finger Recognition task, the ANOVA on reaction times showed a main effect of GROUP [F(3) = 11.995, p < 0.001, η² = 0.142] and CONDITION [F(2) = 36.832, p < 0.001, η² = 0.253], but no GROUP*CONDITION interaction [F(6) = 0.953, p = 0.458, η² = 0.026]. The Inverse condition produced slower reaction times compared to Perpendicular (+62 ± 22 ms p = 0.014, Bonferroni corrected, here and in the following post hoc tests) and Mirror conditions (+184 ± 22 ms: p < 0.001). Slower reaction times were observed in Perpendicular vs. Mirror condition (p < 0.001). Patient groups performed significantly slower than controls: FEP (+137 ± 25 ms, p < 0.001), SCZ (+69 ± 25 ms, p = 0.040), and ASD (+114 ± 25 ms, p < 0.001), and FEP were slower than SCZ (+68 ± 25 ms, p = 0.041) (Figure 1A). Percentage of Correct trials (Supplementary Figure S1) and percentage Coactivation showed similar group differences (Supplementary Figure S2).

Taken together, these results show significant deficits in the Finger Recognition task for patients with FEP, SCZ or ASD compared to control subjects, with generally highest deficits for FEP patients.

In the Rhythm Tapping task, no significant group differences were found in inter-tap-interval (ITI) accuracy [Scheirer-Ray-Hare, H(3) = 7.542, p = 0.056]. Still, ASD patients tended to show a decreased mean ITI under auditory Feedback (40% faster than required), but a better ITI in the NoFeedback condition, resulting in the largest group difference in ITI between Feedback vs. NoFeedback conditions (Supplementary Figure S3).

The three-way ANOVA of Intra Subject Variability (ISV) showed a significant GROUP [F(3) = 11.485, p < 0.001, η² = 0.076] (Figure 2B), CONDITION [F(2) = 247.145, p < 0.001, η² = 0.541], and CUE effect (Feedback or NoFeedback) [F(1) = 182.096, p < 0.001, η² = 0.303], and also a GROUP*CONDITION interaction [F(6) = 2.250, p = 0.038, η² = 0.031]. No GROUP*CUE interaction was found [F(3) = 0.587, p = 0.624, η² = 0.004]. Patients with FEP had a significantly higher average ISV across conditions and cues than all other groups (control p < 0.001; SCZ p < 0.001; ASD p = 0.006, Bonferroni corrected). At 3 Hz, patients with FEP showed a two-fold higher variability in the Feedback condition than the three other groups (Figure 1B).

No significant group differences were found in the Multi Finger Tapping task, nor in the Sequence Tapping task.

In the Line Tracking task, no significant differences were found between Single-and Dual-task conditions in duration, number of velocity peaks, and number of micromovements. Task duration showed significant differences for GROUP [Scheirer-Ray-Hare, H(3) = 22.355, p < 0.001] with FEP and SCZ groups being slower than controls (respectively +5 ± 3 s: p = 0.003 and + 10 ± 3 s: p < 0.001, Dunn’s test), SCZ being also significantly slower than ASD (+5 ± 3 s: p = 0.013) (Figure 1C). Number of velocity peaks also showed significant differences for GROUP [Scheirer-Ray-Hare, H(3) = 18.336, p < 0.001] with FEP and SCZ producing less smooth movement tracking than control (respectively p = 0.010 and p < 0.001, Dunn’s test). Finally, micromovements showed a significant effect of GROUP [Scheirer-Ray-Hare, H(3) = 8.893, p = 0.031] with ASD (Dunn’s test, p = 0.030) and SCZ (p = 0.043) exhibiting more micromovements than controls.

All significant differences presented here remained significant when comparing the FEP group with age-matched controls.

Dexterity impairment profiles: The performances in the four tablet tasks (in Finger Recognition, Rhythm Tapping, Multi Finger Tapping, and Line Tracking) are summarized as radar plots (Figure 2).

No significant correlations were found between age or medication and tablet performance variables used for patient identification (N = 72, all r < 0.261, p > 0.267). No consistent significant correlations were found between tablet measures and clinical scores across the 4 groups (or within the 3 patient groups for PANSS or BPRS).

The control group had significantly lower NSS total scores compared to the 3 patient groups (Figure 1D). Significantly higher scores were found in NSS subscales for the SCZ and ASD group compared to controls (Supplementary Figure S5). NSS (total or subscale) scores were not significantly different between patient groups. Patients with FEP and those with SCZ has similar PANSS scores (Total, Positive, and Negative; Supplementary Figure S7). No correlations were found between NSS scores and tablet measures.

Three measures, showing the most significant differences between FEP and control subjects, were used for calculating ROC curves: Reaction Time and Coactivations of the Finger Recognition task, and ISV of the Rhythm Tapping task. Tablet variables showed similar sensitivity, but better specificity and larger area under the curve than NSS scores (Table 2), whether for One-vs-One (Figures 3A,B) or One-vs-All approaches (Figures 3C,D). Similar results were found when comparing the FEP group against the SCZ group (Supplementary Figure S8).

The CART decision tree (Figure 4) showed a fair classification of subjects, particularly for FEP patients in the first iteration with 13/20 subjects correctly classified using the Rhythm Tapping task. Further iterations allowed the identification of 25/29 controls, 18/20 FEP patients, 18/20 ASD, but only 13/20 SCZ patients.

The complementary Random Forest analysis (Table 3) provided improved results: Good sensitivity (>92%) and specificity (>72%) for classification of control subjects vs. FEP patients. Classification of patients with SCZ or ASD showed high specificity (>90%), but low sensitivity (<38%). The model shows a 100% positive predictive value for ASD and between 50 and 70% for the other groups, while the negative predictive values scale better (between 80 and 100%), specifically for the FEP group with no false negative result. The corresponding confusion matrix from the validation sample (30% of subjects, N = 30) showed that the model predicted correctly 11/12 control subjects, 4/4 FEP, 3/8 SCZ, and 1/6 ASD patients (Table 3).

No significant group differences were found in cortical excitation measures, i.e., for resting Motor Threshold [ANOVA, F(3) = 1.648, p = 0.186] and slope of Recruitment Curve [ANOVA, F(3) = 1.597, p = 0.198]. Detailed results in Supplementary Table S1.

Regarding motor cortex inhibition, ANOVA of SICI showed a significant effect of GROUP [F(3) = 8.791, p < 0.001]. Post-hoc showed significantly lower SICI (40%) in FEP patients (Figure 5A), compared to control subjects (75%, p < 0.001), to SCZ (60%, p = 0.025) and ASD patients (63%, p = 0.009). SICI was similar in SCZ (p = 0.271) and ASD (p = 0.498) compared to controls. When controlling for antipsychotic medication, the difference in SICI between FEP and SCZ remained significant (ANCOVA, F = 5.89, p = 0.021).

Cerebellar inhibition (CBI) did not show any significant effect of GROUP [ANOVA, F(3) = 2.433, p = 0.073]. Nonetheless (Figure 5B), we found weakest CBI in FEP (0% = no CBI) compared to SCZ (3%), ASD (4%) and controls (10%).

Using direct comparison via ROC curves, two tablet performance variables (from the Rhythm Tapping task and Finger Recognition task) discriminated FEP patients from control subjects with similar sensitivity (~75%), but higher specificity (~90%) compared to clinical NSS scores. Better specificity was particularly clear in distinguishing FEP from Control, SCZ and ASD groups (specificity: NSS = 22% vs. dexterity>70%). The One-vs-All approach used here allowed distinguishing between FEP and pooled Control/SCZ/ASD groups. Good discrimination was also found between FEP and SCZ groups (Supplementary Figure S8). One-vs-all comparison was undertaken to test the tablet in an ecological manner, with the rationale of detecting FEP against a population with large variability, i.e., approaching that of the general population, in order to use the tablet as a screening instrument for detection of at-risk population.

Decision tree and Random forest analysis, that use (in contrast to ROC curves) multiple tablet performance measures, were applied to attempt distinguishing subjects of each group. Detection of FEP patients through these methods corroborated our ROC results: according to the positive and negative prediction values and balanced accuracy, patients with FEP were distinguished with least error, followed by Control subjects and patients with SCZ. Classification of patients with ASD was most difficult. Thus Decision tree and Random Forest analysis allow for partial discrimination between all four groups. Although promising, the model requires larger samples to be improved. Together, quantitative tablet-derived measures of manual dexterity provide better behavioral diagnostic markers of FEP than clinical NSS.

Rapid, easy-to-use, quantitative and non-subjective tools remain a key challenge in psychiatry (54). Several approaches have been investigated to identify SCZ and/or FEP by symptom/neuropsychological criteria, by other behavioral abnormalities (e.g., gaze), by cognitive deficits (attention, language), or by biological markers (brain imaging, genotype) (55). It seems therefore worthwhile to compare the detection accuracy of the tablet markers against other markers.

First, symptom criteria: Clinical scales can detect FEP with ~77% sensitivity and ~ 70% specificity among new referrals to mental health services (56), though patients with mild symptoms remain typically undetected (57). Second, other behavioral markers: gaze abnormalities, e.g., antisaccades or number of fixations (58–63), show comparable FEP detection accuracy (~77% sensitivity, ~81% specificity). Third, cognitive abilities: abnormal language production detects FEP patients with sensitivity and specificity of about 70–80% (64, 65). Our dexterity measures thus achieved similar or better sensitivity and specificity than most other symptomatological/behavioral/cognitive approaches. However, none of these markers achieves sufficient detection accuracy on its own, but they may be useful in early screening, which would need complementary confirmation through more costly, but typically more accurate biological markers, whether molecular (66) or brain imaging (55, 67–69). More generally, a multimodal approach to FEP detection seems to be warranted.

As expected, patients with FEP or SCZ showed deficits in tablet performance, in line with previous studies on sensorimotor deficits assessed by NSS in SCZ (18, 43, 70, 71) and in Ultra-High-Risk individuals (41). We further confirmed (Figure 2) differential group deficits in manual dexterity in patients with ASD (13). More specifically, weaker performance in Finger Recognition is consistent with slower and less accurate mental rotation in SCZ (72, 73) and FEP (74), but not in ASD (75, 76). In Rhythm Tapping, only FEP patients were impaired, in contrast to deficient temporal perception shown in SCZ and ASD (77, 78), a discrepancy likely due to task differences. Sensorimotor accuracy in Line Tracking was affected in SCZ and FEP groups, consistent with previous findings (30, 79), whereas ASD patients showed good tracking performance, unlike previously reported using grip force control measures (13). Expected group differences in divided attention, operationalized by the Line Tracking Dual-task condition, were not found, likely due to sub-optimal tablet task conditions.

Although there were no significant group differences in Cerebellar Brain Inhibition, the non-significant trend for reduced CBI in FEP vs. controls is in line with previous reports in SCZ (33).

However, significantly reduced SICI in FEP, and trends to reduced SICI in ASD and SCZ, were consistent with previous results in FEP (44), ASD (80–82), and SCZ (30, 83, 84). Age disparity (younger FEP subjects) is not likely to explain this SICI group difference since SICI was shown to not relate to age (85). We found no evidence for altered cortical excitability in patient groups, in line with our previous report (30), but in contrast to other studies (86, 87). Together these results suggest that reduced SICI is a stronger marker for FEP, a neurodevelopmentally volatile phase (5), than for SCZ or ASD. This needs confirmation in larger studies and might be expanded to Ultra-High-Risk patients.

Here we discuss potential issues that may have affected our data and biased the results. This was a pilot study and our results need to be confirmed in larger samples. Expansion to ultra-high risk (UHR) subjects could provide valuable insights to sensorimotor signatures present early in the development of psychosis. Regarding the measures used, the (18) NSS scale, associated with structural motor cortex alterations (70), is not the only available NSS scale. We do not know whether other NSS scales would have given similar results. Subclinical attentional deficits may have affected our dexterity measures (30). We attempted (without success) to quantify divided attention through the dual-task condition in the line-tracking task. Age difference across groups, unavoidable in comparisons of FEP against the other groups, may have contaminated results. However, we did find a similar difference in tablet performance in FEP when comparing to an age-matched sub-sample. Another potential methodological bias was that we added 9 control subjects in the classification analysis. However, FEP remained the best classified group even with 20 subjects in each group. Medication did not seem to confound our results since including medication in the ANCOVA analysis gave similar results. We did not have information on whether patients had received psychotherapy, which can improve symptoms and functioning of psychotic patients (88, 89). Regarding the interpretation of results, one could also argue that we did not distinguish FEP from SCZ and ASD, but only discriminated between acute psychotic disorder (FEP) and chronic psychotic disorder [as in SCZ and ASD; (4)]. However, patients with FEP and those with SCZ had similar PANSS scores (see Supplementary Figure S7), suggesting that our measures of dexterity capture sensorimotor aspects particular to FEP and not explained by severity of psychotic symptoms. Finally, in our TMS measurements of CBI we used rather low and constant stimulation intensity (to avoid discomfort) and CBI may have been stronger if using higher stimulation intensities (51).

Patients with first-episode psychosis showed impairments in manual dexterity compared to control subjects when assessed by a new tablet application. Differential manual deficits were also present in patients with schizophrenia or autism spectrum disorder. Tablet-based measures, used as behavioral markers, allowed identification of first-episode psychosis patients vs. control subjects, and differentiated FEP from patients with schizophrenia and patients with autism spectrum disorder with good sensitivity and specificity, and more accurately than detection using neurological soft signs. Neurophysiological alterations were also present with a strongly reduced cortical inhibition in patients with first-episode psychosis. The behavioral results suggest that tablet-based measures of manual dexterity may serve as promising complementary markers for the detection of first-episode psychosis. Replication in a larger cohort is currently under investigation.