Search strategy for studies
We searched Medline (from 1966), CINAHL (from 1982), Embase (from 1980), and PsycINFO (from 1965) to December 2003. We used all database controlled vocabulary headings for conversion disorder and hysteria and the text words psychosomatic, psychogenic, somatization, unexplained, conversion, non-organic, dissoc*. They were combined with text words for paralysis, paresis, sensory disturbance, deafness, hearing, vis*, blind*, and movement disorders. References for pseudoseizures were searched with the text words: pseudoseizure, non-epileptic, psychogenic seizure, hysterical attack. All references under the heading “conversion disorder” or with the text word hysteri* were also examined. We reviewed the titles and abstracts online and obtained copies of all publications that might conceivably contain relevant data. The reference lists of all these publications were also examined for additional relevant studies published after 1965.
Study inclusion and exclusion
We included studies if the participants were aged > 16; symptoms were described as medically unexplained, non-organic, psychogenic, hysterical, conversion, or functional; the symptoms described were motor (paresis, paralysis, movement disorder, gait disorder), sensory (numbness or paraesthesia), loss of vision, loss of hearing, or episodes resembling epilepsy (pseudoseizures); the study was of more than 10 patients; and there was a follow-up period of more than six months, at which time some attempt was made to review the accuracy of the initial diagnosis. We excluded studies of patients with other somatoform diagnoses including somatoform pain disorder and somatisation disorder (multiple chronic symptoms unexplained by a general medical condition attributable to several bodily systems). No studies were excluded on the basis of language.
We considered that a misdiagnosis of conversion disorder had occurred when the investigators concluded that, with hindsight, most of a patient's original symptoms or signs were better explained by a disease. We used the term “disease” to describe a clearly defined pathology (for example, stroke) or a diagnosis that is generally accepted as a medical condition (for example, migraine, dystonia). We did not record a misdiagnosis if the patient had an initial diagnosis of disease with a comorbid diagnosis of “hysteria” that was subsequently revised to a diagnosis of disease alone. Neither did we record psychiatric misdiagnosis. A symptom such as leg paralysis cannot in hindsight be attributed to another psychiatric diagnosis such as depression even if comorbid depression was missed at the time of initial diagnosis.
Data extraction and analysis
Four investigators (JS, RS, AC, and MS) independently reviewed all reports and a fifth (CW) arbitrated in cases of disagreement. We collected data on the nature of the symptoms; the sex and mean age of the participants; the setting in which the patient was seen; whether sampling was consecutive or non-consecutive, and retrospective or prospective; the years in which the patients initially received the diagnosis; the duration and completeness of follow-up; the method of diagnosis used at follow-up; and the frequency and nature of any misdiagnoses and causes of death (where this was recorded). We attempted to contact authors to clarify data when these were uncertain.
We calculated the rate of misdiagnosis in each study as the “number of patients misdiagnosed” divided by the “number of patients followed up including those who died.” Firstly, we determined an overall rate of misdiagnosis both by simple pooling of the data and by a random effects model. Secondly, we summarised the data according to the date the initial “conversion” diagnosis was made (rather than study publication date). If the dates of recruitment of patients were unavailable we estimated them using the published mean or range of duration of follow-up (allowing one extra year for publication). We summarised these data by charting individual studies according to the midpoint of their recruitment period, and, because this may not reflect adequately the wide variation in the duration of patient recruitment, we also calculated the proportion of patients with a misdiagnosis for each five year and ten year time period from 1950-99 using random effect models. Thirdly, we used a general linear mixed model to examine the relation between the proportion of patients with a misdiagnosis and the variables of age, sex, duration of follow-up, and midpoint of study recruitment. All analyses were performed with study effects fitted as random, using PROC GLIMMIX for SAS 9 (SAS Institute, Cary, NC).