Feigning spectrum behaviour on the Neck Disability Index and Impact of Events Scale in whiplash associated disorder after motor vehicle crashes: a systematic assessment of classification models.

purposeThe aim of this study was to develop indices of feigning spectrum behaviour (FSB) on the Visual Analogue Pain Scale (VAS) Neck Disability Index (NDI) and Impact of Events Scale (IES) in people with whiplash associated disorder (WAD) after motor vehicle crashes (MVC) using a "known group" or "criterion validity" design. This study represents preliminary steps to achieving this aim.

methods251 MVC injured people opted into a whiplash clinic program and completed the Test of Memory Malingering (TOMM), VAS, NDI, and IES. Participants gave their consent to participate in the research and a de-identified data set was supplied to the researchers. The TOMM cut score of below 45 on Trial 2 was used to classify FSB-Group (FSBG) and Genuine Responding (GR) for comparison. A cut score identified by Pina et al of above 29 on the NDI was used to classify Genuine Responders-Pina (GR-P) and FSB-Pina (FSB-P) for comparison.

resultsThe majority of the participants were in the acute phase of WAD (Mean = 6.74-weeks, SD = 5.30). 29 of the participants were classified as FSBG and 63 were classified as FSB-P. The FSBG scored significantly higher (sig. >0.001) on the VAS, NDI and IES. Statistical and machine learning (ML) classification methods were systematically compared. Receiver Operator Characteristics showed a cut score of 37 and above on the NDI had an acceptable Specificity of 95.5% and a Sensitivity to feigning of 34.5%. Similarly, a cut score of 62 and above on the IES demonstrated an acceptable Specificity of 95.5% and a Sensitivity to feigning of 27.6%. The best ML models were Classification and Regression Tree (CRT) models based on the item level responses of the NDI correctly classified 94.8 % of cases, identifying 100% of GR and 55.2% of FSB, and item level responses on IES correctly classifying 93.6% of GR-P and 73% of FSB-P. 5-Fold Cross validation revealed overfitting on the IES CRT for GR-P and FSB-P but not models for FSBG and GR. A cut scores of 80.5mm on the VAS and 56 on IES correctly classified 38.1% and 34.9% of FSB-P with 90% specificity for GR-P.

conclusionCut scores on the NDI and IES were computed, and statistical models were systematically generated to distinguish between GR with good specificity and a modest degree of sensitivity to FSB. Overall, Decision Trees out-performed other ML models with the NDI being more sensitive to FSB than the IES, or when both measures were combined. Qualitatively, questions involving symptoms interfering with concentration, work, pain intensity on the NDI, and repressing thoughts about the MVC appear to best distinguish between GR and FSB.