Latent class analysis of co-morbidity in the Adult Psychiatric Morbidity Survey in England 2007: implications for DSM-5 and ICD-11

Abstract

BackgroundPsychiatric co-morbidity is complex and ubiquitous. Our aim was to describe the extent, nature and patterning of psychiatric co-morbidity within a representative sample of the adult population of England, using latent class analysis.MethodData were used from the 2007 Adult Psychiatric Morbidity Survey, a two-phase national household survey undertaken in 2007 comprising 7325 participants aged 16 years and older living in private households in England. The presence of 15 common mental health and behavioural problems was ascertained using standardized clinical and validated self-report measures, including three anxiety disorders, depressive episode, mixed anxiety depressive disorder, psychosis, antisocial and borderline personality disorders, eating disorders, post-traumatic stress disorder, attention deficit disorder, alcohol and drug dependencies, problem gambling and attempted suicide.ResultsA four-class model provided the most parsimonious and informative explanation of the data. Most participants (81.6%) were assigned to a non-symptomatic or ‘Unaffected’ class. The remainder were classified into three qualitatively different symptomatic classes: ‘Co-thymia’ (12.4%), ‘Highly Co-morbid’ (5.0%) and ‘Addictions’ (1.0%). Classes differed in mean numbers of conditions and impairments in social functioning, and these dimensions were correlated.ConclusionsOur findings confirm that mental disorders typically co-occur and are concentrated in a relatively small number of individuals. Conditions associated with the highest levels of disability, mortality and cost – psychosis, suicidality and personality disorders – are often co-morbid with more common conditions. This needs to be recognized when planning services and when considering aetiology.

Latent class profile of psychiatric symptoms and treatment utilization in a sample of patients with co-occurring disorders

“…These results are in line with a body of work which points toward the integration of both a categorical and a dimensional (understood as symptom severity, expressed by the score of a particular subscale item) approach to patient classification. For instance, studies seeking to find subgroups based on combinations of psychiatric comorbidities in adolescents, 21 patients with posttraumatic stress disorder (PTSD), 22 patients with schizophrenia, 23 and the general population [24][25][26] obtained solutions with quantitative and qualitative differences between classes, suggesting that subgroups are mostly based on combinations of specific disorders and symptom severity. In another study, 27 carried out with adults seeking treatment for substance use, the best fit was obtained by a three-class model with quantitative differences only (classes were labeled as SUD-only, co-occurring major depressive disorder, and multimorbidity).…”

Specificity of childhood psychotic symptoms for predicting schizophrenia by 38 years of age: a birth cohort study

“…This is in keeping with other recent studies (Kelleher et al 2012b;Werbeloff et al 2012) and commentaries (Lin et al 2012;, which also suggest that it may no longer be useful to extrapolate findings based on early psychotic symptoms solely to psychotic disorders, as it can no longer be assumed that individuals with these symptoms will eventually develop a psychotic disorder. Indeed, the lack of specificity of these early psychotic symptoms should not be that surprising because in childhood they co-occur with other mental health problems (including self-harm; Polanczyk et al 2010) and both prodromal (Kelleher et al 2012d) and clinical schizophrenia (Murray et al 2003;Weich et al 2011) have high rates of co-morbidity.…”

Associations between sheep farmer attitudes, beliefs, emotions and personality, and their barriers to uptake of best practice: The example of footrot

“…To increase confidence that the final solution for each model had converged on the global maximum solution, models were repeatedly estimated with increasing random start values until the log likelihood was replicated several times (Nylund et al, 2007). Goodness-of-fit statistics (AIC, BIC, Entropy, LMR, BLRT) as outlined by (Weich et al, 2011) combined with intuitive reasoning were used to select the final latent class model. Upon reaching a final solution, the posterior probability for each farmer being in each class and then the conditional probability that farmers in a class were practising a management were calculated.…”