Register-Based Cohort Study

A register-based cohort study was conducted on children who was resident in Stockholm County in Sweden. They were followed up from 2001 to 2007. In order to identify the mental illness in parents and eating disorders in their children, researchers used data from several different databases and registers.
In addition, From database and registers, they also gathered other information that might affect the study result, such as mother’s age at child’s birth, educational level, and occupation class. These info was regarded as potential confounders which will be adjusted in the data analysis. To the figure on the right, on the top is the total raw sample size, they ruled out children who were not native born, or those who had parents born outside

of Sweden. Then they removed those with missing data on potential confounders. Finally,they had a final study sample.
With the Final study sample and eating disorder and mental illness information, they started the statistical analysis. At first, they used Cox proportional hazards regression to estimate the hazard ratio, the corresponding 95% confidence interval of eating disorder and p value. Second, they used multivariable analysis to adjust for factors which might confound the relationship. As a result, they got an adjusted hazard ratio, a new confidence interval and p value.
Let’s take a look at the study result. First, we need understand what is hazard ratio. I got this long definition from wikipedia. To make it easier to understand, i will give an example: If there is a 2.0 hazard ratio, it means the children will have twice of the chance to get eating disorders. In addition, we also need to understand the concept of confidence interval and p value to know how significant is the data. Again, to make things easy, if the p value is less than 0.05, it means the hazard ratio is trustworth. Now let’s look at the the bottom table, the last column gives us the adjusted hazard ratio of different parental mental illness. Even though all the mental illnesses give a adjusted hazard ratio larger than at least 1.25, but only bipolar disorder, personality disorder and depression/anxiety give a p value smaller than 0.05.