Development of the CAT–FER: A Computerized Adaptive Test of Facial Emotion Recognition for Adults With Schizophrenia

Development of a Rasch-calibrated emotion recognition video test for patients with schizophrenia

Patients with schizophrenia tend to have deficits in emotion recognition (ER) that affect their social function. However, the commonly-used ER measures appear incomprehensive, unreliable and invalid, making it difficult to comprehensively evaluate ER. The purposes of this study were to develop the Computerized Emotion Recognition Video Test (CERVT) evaluating ER ability in patients with schizophrenia. This study was divided into two phases. First, we selected candidate CERVT items/videos of 8 basic emotion domains from a published database. Second, we validated the selected CERVT items using Rasch analysis. Finally, the 269 patients and 177 healthy adults were recruited to ensure the participants had diverse abilities. After the removal of 21 misfit (infit or outfit mean square > 1.4) items and adjustment of the item difficulties of the 26 items with severe differential item functioning, the remaining 217 items were finalized as the CERVT items. All the CERVT items showed good model fits with small eigenvalues (≤ 2) based on the residual-based principal components analysis for each domain, supporting the unidimensionality of these items. The 8 domains of the CERVT had good to excellent reliabilities (average Rasch reliabilities = 0.84-0.93). The CERVT contains items of the 8 basic emotions with individualized scores. Moreover, the CERVT showed acceptable reliability and validity, and the scores were not affected by examinees' gender. Thus, the CERVT has the potential to provide a comprehensive, reliable, valid, and gender-unbiased assessment of ER for patients with schizophrenia.

Psychometric Properties of Two Tasks in the Allen Cognitive Level Screen-Sixth Edition for Community-Dwelling People Living With Schizophrenia

IMPORTANCE

Empirical evidence is needed on the psychometric properties of the Allen Cognitive Level Screen-Sixth Edition (ACLS-6), an instrument that assesses cognitive functions and is commonly used for people living with schizophrenia.

OBJECTIVE

To examine the convergent validity, discriminative validity, and test-retest reliability of two tasks, stitching and copying, in the ACLS-6 for community-dwelling people living with schizophrenia.

DESIGN

Prospective observational study.

SETTING

Psychiatric center.

PARTICIPANTS

110 people living with schizophrenia.

OUTCOMES AND MEASURES

To examine convergent validity, we calculated correlations (Pearson's r) between the two tasks and between these two tasks and three cognitive measures. We checked for floor and ceiling effects and conducted independent t tests to evaluate discriminative validity. We calculated intraclass correlation coefficients (ICCs) to investigate test-retest reliability.

RESULTS

We found a strong correlation (r = .88) between the two tasks and moderate correlations (rs = .32-.52) between the two tasks and the three cognitive measures. No floor or ceiling effects were observed for the two tasks, and t tests showed significant differences between two participant groups with marginal and mild clinical symptoms (p < .001). The ICC values for the two tasks were .71-.74.

CONCLUSION AND RELEVANCE

The stitching and copying tasks of the ACLS-6 have good convergent validity, discriminative validity, and test-retest reliability for community-dwelling people living with schizophrenia. The copying task showed a strong correlation with the stitching task and a similar score range, so practitioners can consider using the copying task as a substitute for the stitching task. What This Article Adds: The stitching and copying tasks of the ACLS-6 have sound psychometric properties for measuring cognitive functions in community-dwelling people living with schizophrenia.

Error patterns of facial emotion recognition in patients with schizophrenia

Error patterns of facial emotion recognition (FER) indicate how individuals misinterpret others' facial expressions, which helps clinicians to manage related deficits. However, previous investigations are limited and may have been biased due to methodological issues (e.g., no consideration of response bias). This study aimed to propose a detectability index (d') for adjusting response bias and examine the error patterns of FER in patients with schizophrenia. Responses to 168 photos showing seven basic emotions, obtained from 351 patients with schizophrenia and 101 healthy adults, were extracted from a previous study. The differences in the d's between the two groups (Δd') were calculated to examine the error patterns of FER among the seven emotions. The findings were generally overlapped with those identified by the traditional confusion matrix. Four error patterns were found. First, the patients were insensitive to some negative emotions (i.e., sadness [Δd' = 0.83] and fear [Δd' = 0.72]). Second, they misrecognized happy faces as showing negative emotions (e.g., disgust [Δd' = 0.43] and sadness [Δd' = 0.37]). Third, they misinterpreted surprised faces as all the other emotions (Δd' = 0.41-0.87), except neutral. Fourth, they confused some negative emotions (e.g., misrecognizing fear as anger [Δd' = 0.87]). Our findings suggest that patients with schizophrenia show four error patterns of FER compared to healthy adults. Accordingly, interventions could be selected to improve their sensitivity to faces with negative emotions, differentiation of faces among positive and negative emotions, understanding of surprised faces, and discrimination of faces with negative emotions.

Using machine learning to improve the discriminative power of the FERD screener in classifying patients with schizophrenia and healthy adults

Background Facial emotion recognition deficit (FERD) seems to be an obvious feature of patients with schizophrenia and has great potential for classifying patients and non-patients. The FERD screener was previously developed to classify patients from healthy adults. However, an obvious drawback of this screener is that the recommended cut-off scores could enhance either sensitivity or specificity (about 0.92) only, while the other one is at an only acceptable level (about 0.66). Machine learning (ML) algorithms are famous for their feature extraction and data classification abilities, which are promising for improving the discriminative power of screeners. This study aimed to improve the discriminative power of the FERD screener using an ML algorithm. Methods The data were extracted from a previous study. Artificial neural networks were generated to estimate the probability of being a patient with schizophrenia or a healthy adult based on the examinee's responses on the FERD screener (168 items). The performances of the ML-FERD screener were examined using a stratified five-fold cross-validation method. Results Across the five subsets of data, the ML-FERD screener showed extremely high areas under the receiver operating characteristic curve of 0.97-0.99. With the optimized cut-off scores, the average sensitivity and specificity of the ML-FERD screener were 0.90 (0.85-0.93) and 0.93 (0.86-1.00), respectively. Limitations The characteristics of patients were not representative, and the age was mismatched to control group. Conclusion The ML-FERD screener appears to have a better discriminative power to classify patients with schizophrenia and healthy adults than does the FERD screener.

Sensitivity and specificity of a facial emotion recognition test in classifying patients with schizophrenia

BACKGROUND

Schizophrenia is a debilitating mental illness that causes significant disability. However, the lack of evidence for functional decline yields difficulty in distinguishing patients with schizophrenia from healthy adults. Since patients with schizophrenia demonstrate severe facial emotion recognition deficit (FERD), FERD measurement appears to be a promising solution for the aforementioned challenge.We aimed to develop a FERD-based screening tool to differentiates patients with schizophrenia from healthy adults.

METHODS

Patients' responses were extracted from a previous study. The most discriminative index was determined by comparing the area under the receiver operating characteristic curve (AUC) of patients' FER scores in 7 domains individually and collectively. The best cut-off score was selected only for the most discriminative index to provide both high sensitivity and specificity (≥ 0.90).

RESULTS

The "number of domains failed" showed the highest discriminative value (AUC = 0.92). Since high sensitivity and specificity could not be achieved simultaneously, two sub-optimal cut-off scores were recommended for prospective users. For users prioritizing sensitivity, the "≥ 2 domains failed" index yields high sensitivity (0.96) with modest specificity (0.66). For users targeting specificity, the "≥ 4 domains failed" indexachieves high specificity (0.92) with acceptable sensitivity (0.72).

LIMITATIONS

Convenience sampling with mild clinical severity and younger healthy adults (< 20 years old) may limit the generalizability.

CONCLUSION

The FERD screener seems to be a discriminative tool with changeable cut-off scores achieving high sensitivity or specificity. Therefore, it may be useful in detecting patients and ruling out adults erroneously suspected of having schizophrenia.