An account is needed to access this study, but it can be found at http://www.medscape.com/psychiatryGraphology for the Diagnosis of Suicide Attempts: a Blind Proof of Principle Controlled Study
S. Mouly; I. Mahé; K. Champion; C. Bertin; P. Popper; D. De Noblet; J. F. Bergmann
Int J Clin Pract. 2007;61(3):411-415. ©2007 Blackwell Publishing
Posted 04/18/2007
Summary and Introduction
Summary
To evaluate the ability of two graphologists and two practising internists not trained in graphology to differentiate letters written by subjects who have attempted to commit suicide by self-poisoning and healthy volunteers, we performed a maximal blind controlled study vs. healthy volunteers. Forty fully recovered patients who had attempted to commit suicide and 40 healthy volunteers wrote and signed a short letter or story not related to the parasuicide or their mental health status. The evaluators classified the 80 letters as 'suicide' or 'no suicide' in an intention-to-treat analysis. Letters expressing sadness were subsequently excluded for a per-protocol analysis. Correct diagnosis of suicide and of healthy controls was made in, respectively, 32 of 40 and 33 of 40 letters by the graphologists and in 27 of 40 and 34 of 40 letters by the internists. After the exclusion of 12 letters expressing sadness, the sensitivity, specificity, positive predictive value and negative predictive value were, respectively, 73, 88, 81 and 82% for the graphologists and 53, 89, 80 and 71% for the internists. Both classified the letters with significantly more effectiveness than chance (p < 0.001) with no statistically significant difference between the two groups of evaluators. We concluded that graphological analysis is able to differentiate letters written by patients who attempt suicide from those written by healthy controls. This technique shows an acceptable degree of accuracy and could therefore become an additional discharge or decision-making tool in Psychiatry or Internal Medicine.
Introduction
Suicide is a major public health problem, but there is little evidence regarding the effectiveness of prevention programmes.[1] Predictive evaluation of the risk of suicide is difficult both in general practice[2] and in psychiatric units.[3] The level of self-control in a patient presenting risk factors for suicide cannot be evaluated using a single test,[4] but it is nevertheless important that this should be approached for therapeutic decision-making, hospitalisation or discharge.
Medical graphology is a subjective method of handwriting analysis for evaluation of personality.[5] But only few psychological characteristics such as psychoses, psychosomatic symptoms or extraversions[6] have been correlated with graphological characteristics. Because of the poor methodology of these studies, it is not clear whether graphology is able to determine the psychological profile of patients.[7] Graphologists claim to be able to determine pathological behaviour such as hysteria or depression by the simple analysis of a single 'neutral' letter from patients and to be able to exclude these diagnoses in a healthy control population. But the sensitivity and specificity of graphology in the diagnosis of psychiatric diseases have never been properly evaluated, in particular because of the difficulty in obtaining a definitive positive diagnosis in patients and excluding such a diagnosis in healthy controls.[8] For graphologists, handwriting analysis can determine suicidal behaviour before or after a suicide attempt. Letters from patients recovering from a recent suicide attempt are thus a good model for evaluating graphology in a controlled study vs. letters from healthy subjects without a history of suicide. If this method proves to be efficient for a psychiatric diagnosis of suicide, it could be proposed as a part of the clinical evaluation of suicide risk in patients.
Methods
We asked consecutive patients hospitalised in the intensive care unit of two large primary care Parisian hospitals for deliberate self-poisoning with drugs to write a story or a short letter relating a childhood memory not linked to the parasuicide. This request was made on the day of discharge from the hospital to patients free of any prescribed drug who were alert and who had undergone a normal neurological clinical examination. The patients were given sheets of white paper and various writing tools (pen, ball-point pen and pencil). The letter had to be signed. The control group consisted of healthy volunteers with no history of parasuicide and no depression evaluated by a MADRS (range 0-60) below 16. Patients and controls received written information about the study and gave written informed consent. A randomised number was given to each letter by an investigator (CB) not participating in the graphological evaluation. The letters from the patients and controls were then mixed. Information concerning age, sex and left or right hand status was given with the letters to two independent teams of evaluators blind to the diagnosis: (i) two graphologists and (ii) two physicians from an internal medicine unit without any knowledge about graphology. The four evaluators did not know any of the patients or the control subjects. Each evaluator decided to dichotomously classify each letter as suicide or control. In the event of a discordance between the two graphologists or between the two internists, an attempt to reach a consensual diagnosis was made by means of an open discussion.
We calculated that a sample size of 80 (40 per group) was necessary to observe a difference of 20% between the graphologists and chance (correct diagnosis by chance 50%, correct diagnosis by graphologists 70%) with β = 20%, α = 5%. Intention-to-treat analysis included the 80 letters. Per-protocol analysis was made after the exclusion, by an investigator blind to the diagnosis (JFB), of the letter expressing sadness which would have orientated the final diagnosis. The χ2 one-sample test was used to compare the observed results with a chance response of 50%.
Results
The letters were collected between April and September 2002. The two suicide (n = 40) and control (n = 40) groups were similar in terms of mean age (respectively 38.0 and 37.7), sex (male 12 and 15) and left hand status.[3,1] The mean MADRS score in the control healthy group was 6.72 ± 3.7.
Classification of the letters as suicide or controls by the graphologists and internists is presented in Table 1 . All 80 letters were included in the ITT analysis. A discordance between the two graphologists was observed in 12 cases, but the consensus discussion finally led to correct diagnosis in eight cases. In brief, the graphological diagnosis had a sensitivity of 80% (IC 66-90), a specificity of 82% (IC 95%: 67-91), a positive predictive value of 82% (IC 95%: 67-91) and a negative predictive value of 80% (IC 95%: 66-90). Their final diagnosis was statistically superior to a chance distribution (α2 = 15.6, df = 2, p < 0.001). For the internists, the results were sensitivity 0.67, specificity 0.85, positive predictive value 0.82 and negative predictive value 0.72, also statistically different to chance (χ2 = 12.2, df = 2, p = 0.005). There was no statistical difference regarding the accuracy of the diagnosis between the graphologists and internists (p = 0.45).
For the per-protocol (PP) analysis, 12 letters expressing sadness were excluded: 10 from the suicide group and two from the control group. Eleven of these were classified as 'suicide' by both teams of evaluators in the ITT analysis. In the PP analysis, the characteristics of the graphological test for graphologists and internists after confrontation were, respectively, sensitivity 73 and 53%, specificity 88 and 89%, positive predictive value 81 and 80%, negative predictive value 82 and 71%. Both the graphologists and the internists remained significantly different to chance in the PP analysis (p < 0.001 for both evaluator groups). After initial evaluation of the 80 letters, the two graphologists disagreed on the diagnosis of 12 letters; the consensus session led to a final correct diagnosis in eight letters. For the internists, diverging opinions were observed in 22 letters and a consensus was obtained with a final correct diagnosis in 14.
Discussion
This pilot study is the first controlled trial to have demonstrated the ability of graphologists to detect a particular psychiatric status, i.e. parasuicide with a sensitivity and specificity around 80%.
Graphology is based on a theory according to which the shape, rhythm, size and position of the handwritten letters and words reflect the character of the writer.[5] But the correlation between the writing and the psychological profile is essentially based on case reports in which graphological characteristics are described with full knowledge of the psychological characteristics of the subject.[9] However, it would be very useful to be able to precisely determine the psychological status of an individual by studying his/her handwriting. If handwriting could define a dangerous situation such as, for example, a suicide risk, and if graphological analysis could predict this risk with a good level of predictive value, it could become a tool in the decision-making process.[10] The decision to hospitalise or discharge a depressive patient is always difficult to take due to the risk of suicide which, if incorrectly assessed, can have serious consequences.[11,12] We therefore chose this parasuicide model in order to try to validate the graphological diagnosis. As graphologists believe that suicidal tendencies continue to be reflected in handwriting after a suicide attempt, it was methodologically justified to study the letters written by subjects who had survived such an attempt. These subjects were known to have attempted suicide and there was no risk of diagnostic error amongst the suicide population. We therefore preferred not to evaluate their degree of depression by asking them to complete a MADRS questionnaire, so that they did not feel the study to be a psychometric test, which might have influenced their writing. We performed the study in patients who had recovered a normal level of alertness, on the day of discharge from the hospital. To keep the study blinded, patients were asked to write a letter not related to their parasuicide or their mental health status or history. The control subjects did not show any depressive or suicidal tendencies, and we believe that the populations studied were correctly chosen for measurement of the metrological properties of graphology. Our findings show that there truly are detectable differences between the writing of subjects who have attempted suicide and healthy subjects even if both graphologists missed the suicidal intent in 23% of cases. The fact that this difference was also detected by practising internists who were not graphologists could represent an argument against the specific and rigorous nature of this technique. But one may also consider that the principles of graphological diagnosis are based on good sense and a logic which is perceptible by nonspecialists. However, the graphologists were slightly though not significantly better than the internists in this study with a sensitivity of 73 and 53%, respectively.
Our study had some limitations. The study evaluated the accuracy of only two graphologists in only one medical situation without follow-up of the patients. Our results could not be extrapolated to all graphologists and to other psychological disorders. The grounds which led the graphologists to consider the letters to have been written by subjects who had attempted suicide are not unequivocal. There is no global or composite score, but factors such as falling lines, a loss of the link between the letters, an heterogeneity of the graphics, a weak holding of the lines, a change in the disposition of the words, an unusual possession of the space of the page are taken into account. The subjective dichotomous global decision was built on the size, shape, pressure, speed and movement of the letters and words and on the signature. However, no graphic element in itself was purely significant of one definite quality leading to a prediction of suicide.
As we did not include a formal depression rating scale to evaluate subjects who had attempted suicide, it was impossible to search for a correlation between handwriting analysis and level of depression. We preferred to keep a more powerful binary primary end-point in accordance with our sample size calculation. A previous study comparing the assessment of 10 graphologists in the diagnosis of extraversion showed good agreement between the graphologists, but in this study, only six subjects were evaluated without any healthy control group.[6] In another study,[8] graphology was compared with results of psychological tests for vocational guidance.
The experimental conditions of our pilot study certainly do not allow definitive conclusions to be drawn as to the validity of graphology. Our design with a dichotomic answer of suicide yes-no greatly facilitates graphological analysis. The content of the letters themselves may influence analysis of the handwriting; indeed, 11 of the 12 letters expressing sadness were classified in the suicide group. However, even after the exclusion of these letters, graphological analysis still had a predictive value of about 80%. Our study should be considered as a preliminary report, a kind of go-no go study. Had our findings been negative, we would have demonstrated, using a solid methodology, that graphology was not reliable in the diagnosis of psychiatric illnesses such as suicide. Conversely, our positive results are not sufficient to provide overall validation of this method. A future study using letters written by foreign patients in a language unknown by the graphologists and/or mixing several different types of clearly differentiated psychiatric illnesses and letters written by healthy volunteers would allow our findings to be clarified. Of course, the prediction of future suicide risk is clinically more important than establishing a previous suicide attempt but needs longer and larger prospective studies and was not the goal of our proof of principle trial.
In conclusion, we have shown that graphological analysis is different to chance in evaluating the presence of a suicide risk in handwriting, but further studies are needed before the true role of this technique in the diagnostic process can be established.
An association between words used to describe handwriting features and personality traits might be a reason for our positive results.[13] A high risk for suicide might modify the form or the position of the letters, the words or the lines in a letter and/or a signature, but the goal of our study was not to attempt to explain the theory of graphology but simply to evaluate the value of graphology as a diagnostic test. Even with 27% of false positives and 13% of false negatives, this study yielded positive results and should lead to further studies. Prospective studies with a follow-up of patients with various graphological diagnoses of psychiatric diseases would allow us to ascertain whether graphology provided additional information to the usual psychological diagnosis. In this case, graphology could act as a complement to standard methods and become an additional tool in therapeutic decision-making.[14] Indeed it should be noted that in psychiatry, many therapeutic decisions are taken on the basis of impressions the predictive value of which has been even less well studied than the graphology in the present study.[15]
We can conclude that graphology is able to differentiate letters written by patients who attempt suicide from those written by healthy subjects with an acceptable degree of accuracy. This difference is not due to the ideas expressed in the content of the letters.
Table 1. Number Of Correct Graphological Diagnoses (%) of Parasuicide by two Graphologists and two Internists After Reading Letters From Patients Following Attempted Suicide (n = 40) and Healthy Controls (n = 40)
Predicting suicide through handwriting?
Moderator: Alyrium Denryle
Predicting suicide through handwriting?
To do a little math with Bayes' Theorem, suppose that 1% of the population under study plans to commit suicide. Out of those, 80% (0.8%) will be correctly identified as suicide patients. Out of the 99% who are not going to commit suicide, 20% (19.9%) will be incorrectly identified as suicide patients. So if someone is identified by this program as a suicide patient, the chance of this being an accurate diagnosis are (0.8%/(19.9% + 0.8%)) = 3.86%. Not very high. Even assuming 10% of patients will commit suicide, the chance of an accurate diagnosis of suicide is 44.4%. And no, you can't repeat this test twice to get a better diagnosis, because this is an handwriting test: presumably if a person's handwriting is tested once, it will test the same way all or most of the time if the test is repeated.