Observational cross-sectional study with the female sample of the UK Biobank cohort. MSc dissertation, part 2.
This observational cross-sectional study aimed to extend the evidence base on cognitive function in women with a history of eating disorders. It hypothesized that women with ED would exhibit a different way of cognitive functioning compared to women without ED. It proposed two directions. First, it compared the cognitive performance of participants who self-reported a history of ED with a comparison group who did not report any such history. Second, within the group of participants who self-reported a history of ED, it compared the cognitive performance of those who met anorexia Body Mass Index (<17.5 kg/m) criteria at the moment of assessment with those who did not meet such criteria. The sample for this study included 260,601 women, from which 347 reported a history of ED. The data was obtained from the UK Biobank. The full text is available for interested readers.
Methods
Each participant attended for a single baseline assessment visit, lasting around two hours.
Computerised touchscreen questionnaires collected information on age, gender, ethnic group, country of origin, educational level, and neighbourhood deprivation. Participants were also asked to complete information on their lifestyle, including smoking habits and alcohol consumption, and on their lifetime history of medical and psychiatric conditions (including eating disorders), and any medication taken around the time of assessment.
A physical examination calculated the Body Mass Index.
The cognitive data were collected from four instruments, which were tools designed by the UK Biobank or adapted versions of standardised tests. All tasks were presented in a computerised touchscreen format, without a trained tester. The cognitive instruments used for this study were: Reasoning task, Reaction Time task, Visuospatial Memory task, and Prospective Memory task.
Data Analysis
The data analysis was carried out with SPSS21 for Windows. Sociodemographic and clinical data from the ED and non-ED groups were summarised and compared. Independent sample t-tests were conducted for scale measures that were continuous and presented a normal distribution, while Chi2 tests were conducted for categorical measures.
To facilitate interpretation across tests, the raw scores for Reasoning, Reaction Time and Visuospatial Memory tasks were transformed into age-adjusted z-scores, within five-year age bands across the whole UK Biobank cohort. The direction of these scores was reversed if necessary, so that a higher z-score represented better performance.
For both sets of analyses, regression models were run on each cognitive test (dependent variable) separately. A simple regression analysis was used to compare the group differences on the cognitive tests without any covariates. A series of multiple regression analyses considered additional covariates (age, education, Townsend Deprivation Index, ethnic group, English-speaking country of birth, smoking and alcohol status, BMI, comorbid neurological and psychiatric conditions, and psychotropic medication) to reduce the role of confounding. The z-scores were analysed using linear regression, and the coefficients could be interpreted as Cohen’s d effect sizes. The Prospective Memory score was a binary 0/1 score, and so this was analysed in its raw form using logistic regression, with coefficients reported as odds ratios.
Diagnostic checks were performed for linearity, homoscedasticity and normality of residuals in the final adjusted linear regression models, all of which were acceptable. Multicollinearity was checked and found to be in the acceptable range (variance inflation factor values all bellow 1.810).
G*Power 3.1.9.2 (Erdfelder, Faul, & Buchner, 1996) was used to estimate statistical power based on the available sample sizes on the cognitive tests. With the largest available sample size of 346 for the ED group and 258997 for the comparison group (Reaction Time task), and the p value threshold set at 0.0125 (two-tailed, taking account of multiple tests across four dependent variables), there was 89% power to detect a group difference in cognitive performance of d=.2 (small effect size), in unadjusted analyses. For other cognitive tests with smaller available samples, there was sufficient power (at least 80%) to detect larger effect sizes (d=.5).
Results
Sociodemographic, lifestyle, physical and medical characteristics of the sample are available on the original paper, as well as the results of the cognitive assessment tasks. .
Results of the regression analyses of the performance of the ED sample in each of the cognitive tasks, considering history of an ED as the independent variable are available on the original paper. The regression analyses were conducted through six models, which added the covariates gradually. The coefficients for the covariates in the fully adjusted models are available on the original paper.
The sample sizes across the six models within each cognitive task reduced by a relatively small amount, indicating minimal impact of missing covariate data in the adjusted models.
The regression analyses showed extremely small effect sizes in the Reasoning task (d=-.072 to .081 across different levels of adjustment) and in the Visuospatial Memory task (d=-.017 to -.066 across different levels of adjustment). The relationship between these two tests and a history of ED was non-significant along the models (all p>.0125).
The ED sample presented mild difficulties in the Reaction Time task, suggesting their processing speed was slower compared to the non-ED sample, although with very small effect sizes (d=-.096 to -.150 across different levels of adjustment). The relationship between the Reaction Time task and a history of ED was statistically significant in the adjusted models until model 5 (all p<.0125); though, in model 6 the relationship became non-significant (p=.074).
Lastly, results in the Prospective Memory task demonstrated that in general the ED sample and the non-ED sample had similar odds of being correct. The ED group showed slight differences, and after adjusting for the covariates, the odds of being correct were 12% lower, but this was non-significant (p=.616).
Results of the regression analyses of the performance within the ED sub-sample in each of the cognitive tasks, considering anorexia BMI criteria as the independent variable are available on the original paper. The regression analyses were conducted through six models, which added the covariates gradually. The coefficients for the covariates in the fully adjusted models are available on the original paper.
The sample sizes across the six models within each cognitive task reduced by a relatively small amount, indicating minimal impact of missing covariate data in the adjusted models.
After adjusting for the covariates, results appeared highly variable. ED participants with anorexia BMI criteria showed small to medium effect sizes in the Reasoning task (d=.2 to .5 across models), Reaction Time task (d=.1 to .2 across models) and Visuospatial Memory task (d=.3 across models). On the contrary, in the Prospective Memory task, the odds of being correct of the ED group with anorexia BMI criteria were around 5% lower (fully adjusted model) than the ED group without the anorexia BMI criteria. Nonetheless, none of these results were significantly related with the BMI (all p>.0125).
Conclusion
The UK Biobank sample of women with a history of ED presented a similar cognitive performance overall compared to women without any such history. Even though, small but statistically significant difficulties in reaction time were found in the ED group, and these correlated with the ED history. This study failed to detect reliably cognitive differences between women who did and did not meet anorexia BMI criteria.
ED are mortal conditions that affect people worldwide. Although the correlation between ED and cognitive impairment was not statistically significant in this community population cohort setting, difficulties are often found in the clinical setting and in the patients’ everyday life. A neuropsychological intervention could increase the understanding of ED and offer new treatment alternatives, it could improve the prognosis of the disorder, and reduce its severity and mortality, promoting the quality of life of those who battle with these disorders.